[{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"This is not an oversight.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"If one is interested in translation, one would probably want to consider show up as a single dictionary word since its semantic interpretation is not trivially derivable from the meanings of show and up.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The correct ambiguity resolution of the syntactic level therefore helps to resolve the morphological one, and vice versa.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"RandLM 0.2 (Talbot and Osborne, 2007) stores large-scale models in less memory using randomized data structures.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"10 Chinese speakers may object to this form, since the suffix f, menD (PL) is usually restricted to.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"2.5 Connectives with scopes.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"(1997; 1999) do have some statistics that show how IdentiFinder performs when the training data is reduced.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"2.1 Reliable Case Resolutions.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Jud ges A G G R ST M 1 M 2 M 3 T1 T2 T3 AG 0.7 0 0.7 0 0 . 4 3 0.4 2 0.6 0 0.6 0 0.6 2 0.5 9 GR 0.9 9 0 . 6 2 0.6 4 0.7 9 0.8 2 0.8 1 0.7 2 ST 0 . 6 4 0.6 7 0.8 0 0.8 4 0.8 2 0.7 4 M1 0.7 7 0.6 9 0.7 1 0.6 9 0.7 0 M2 0.7 2 0.7 3 0.7 1 0.7 0 M3 0.8 9 0.8 7 0.8 0 T1 0.8 8 0.8 2 T2 0.7 8 respectively, the recall and precision.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Figure 3 shows examples of semantic expectations that were learned.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This extends previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and using a simpler training procedure.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"We show translation results for three approaches: the monotone search (MonS), where no word reordering is allowed (Tillmann, 1997), the quasimonotone search (QmS) as presented in this paper and the IBM style (IbmS) search as described in Section 3.2.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Since the parameter and token components will remain fixed throughout experiments, we briefly describe each.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Gather phrases using keywords Now, the keyword with the top TF\/ITF score is selected for each phrase.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Figure 2 shows examples of extracted NE pair instances and their contexts.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The probability distribution that satisfies the above property is the one with the highest entropy.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Taking only the highest frequency rules is much &quot;safer&quot;, as they tend to be very accurate.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"In addition, this formulation results in a dramatic reduction in the number of model parameters thereby, enabling unusually rapid training.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In this domain the major scenarios involve the things they agreed on, rather than the mere fact that they agreed.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We are very grateful to Tony Kroc.h, Michael Pails, Sunil Shende, and Mark Steedman for valuable discussions. formalisms.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Our primary goal is to exploit the resources that are most appropriate for the task at hand, and our secondary goal is to allow for comparison of our models\u2019 performance against previously reported results.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"an event.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This is orthographically represented as 7C.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The following features were used: full-string=x The full string (e.g., for Maury Cooper, full- s tring=Maury_Cooper). contains(x) If the spelling contains more than one word, this feature applies for any words that the string contains (e.g., Maury Cooper contributes two such features, contains (Maury) and contains (Cooper) . allcapl This feature appears if the spelling is a single word which is all capitals (e.g., IBM would contribute this feature). allcap2 This feature appears if the spelling is a single word which is all capitals or full periods, and contains at least one period.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"This is the first set that gives us a fair evaluation of the Bayes models, and the Bayes switching model performs significantly better than its non-parametric counterpart.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"This family represents an attempt to generalize the properties shared by CFG's, HG's, TAG's, and MCTAG's.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Each word is terminated by an arc that represents the transduction between f and the part of speech of that word, weighted with an estimated cost for that word.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This analyzer setting is similar to that of (Cohen and Smith, 2007), and models using it are denoted nohsp, Parser and Grammar We used BitPar (Schmid, 2004), an efficient general purpose parser,10 together with various treebank grammars to parse the input sentences and propose compatible morphological segmentation and syntactic analysis.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The simplest approach involves scoring the various analyses by costs based on word frequency, and picking the lowest cost path; variants of this approach have been described in Chang, Chen, and Chen (1991) and Chang and Chen (1993).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Hence we decided to select ten commentaries to form a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099, for which the entire range of annotation levels was realized, so that experiments with multi-level querying could commence.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Diacritics can also be used to specify grammatical relations such as case and gender.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"9 61.0 44.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Therefore, the number of fine tags varied across languages for our experiments; however, one could as well have fixed the set of HMM states to be a constant across languages, and created one mapping to the universal POS tagset.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Reading the following record\u2019s offset indicates where the block ends.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"On the other hand, we are interested in the application of rhetorical analysis or \u00e2\u0080\u0098discourse parsing\u00e2\u0080\u0099 (3.2 and 3.3), in text generation (3.4), and in exploiting the corpus for the development of improved models of discourse structure (3.5).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Gazdar (1985) argues this is the appropriate analysis of unbounded dependencies in the hypothetical Scandinavian language Norwedish.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Using this encoding scheme, the arc from je to Z in Figure 2 would be assigned the label AuxP\u2191Sb (signifying an AuxP that has been lifted from a Sb).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"A search restriction especially useful for the translation direction from German to English is presented.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"If either case is true, then CFLex reports that the anaphor and candidate might be coreferent.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Word Re-ordering and DP-based Search in Statistical Machine Translation","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"was done by the participants.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The goal of our research was to explore the use of contextual role knowledge for coreference resolution.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Performance typically stabilizes across languages after only a few number of iterations.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In section 5, we then evaluate the entire parsing system by training and evaluating on data from the Prague Dependency Treebank.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The trie data structure is commonly used for language modeling.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR achieved recall in the 4250% range for both domains, with 76% precision overall for terrorism and 87% precision for natural disasters.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For MUC6, for example, Corporate- Suffix-List is made up of ltd., associates, inc., co, corp, ltd, inc, committee, institute, commission, university, plc, airlines, co., corp. and Person-Prefix- List is made up of succeeding, mr., rep., mrs., secretary, sen., says, minister, dr., chairman, ms. . For a token that is in a consecutive sequence of init then a feature Corporate-Suffix is set to 1.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The result of this is shown in Figure 7.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The type of alignment we have considered so far requires the same length for source and target sentence, i.e. I = J. Evidently, this is an unrealistic assumption, therefore we extend the concept of inverted alignments as follows: When adding a new position to the coverage set C, we might generate either \u00c3\u0086 = 0 or \u00c3\u0086 = 1 new target words.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"+ cost(unseen(fm, as desired.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Further, it needs extra pointers in the trie, increasing model size by 40%.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"\u00ce\u00b2 is the shared hyperparameter for the tag assignment prior and word feature multinomials.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"This line of work has been motivated by empirical findings that the standard EM-learned unsupervised HMM does not exhibit sufficient word tag sparsity.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"irL as the product of the probability estimate for i\u00c2\u00a5JJ1l., and the probability estimate just derived for unseen plurals in ir,: p(i\u00c2\u00a51J1l.ir,) p(i\u00c2\u00a51J1l.)p(unseen(f,)).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"There is a sizable literature on Chinese word segmentation: recent reviews include Wang, Su, and Mo (1990) and Wu and Tseng (1993).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"29 \u00e2\u0080\u0094 95.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Evaluation results within sets Table 1 shows the evaluation result based on the number of phrases in a set.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We extend the Matsoukas et al approach in several ways.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Blum and Mitchell 98) go on to give PAC results for learning in the cotraining case.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Table 3 shows BABAR\u00e2\u0080\u0099s performance when the four contextual role knowledge sources are added.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Note that while the standard HMM, has O(K n) emission parameters, our model has O(n) effective parameters.3 Token Component Once HMM parameters (\u00cf\u0086, \u00ce\u00b8) have been drawn, the HMM generates a token-level corpus w in the standard way: P (w, t|\u00cf\u0086, \u00ce\u00b8) = P (T , W , \u00ce\u00b8, \u00cf\u0088, \u00cf\u0086, t, w|\u00ce\u00b1, \u00ce\u00b2) = P (T , W , \u00cf\u0088|\u00ce\u00b2) [Lexicon] \u00ef\u00a3\u00ab n n \u00ef\u00a3\u00ad (w,t)\u00e2\u0088\u0088(w,t) j \u00ef\u00a3\u00b6 P (tj |\u00cf\u0086tj\u00e2\u0088\u00921 )P (wj |tj , \u00ce\u00b8tj )\u00ef\u00a3\u00b8 P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) [Parameter] P (w, t|\u00cf\u0086, \u00ce\u00b8) [Token] We refer to the components on the right hand side as the lexicon, parameter, and token component respectively.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This combination generalizes (2) and (3): we use either at = a to obtain a fixed-weight linear combination, or at = cI(t)\/(cI(t) + 0) to obtain a MAP combination.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Without using the same test corpus, direct comparison is obviously difficult; fortunately, Chang et al. include a list of about 60 sentence fragments that exemplify various categories of performance for their system.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Raj and Whittaker (2003) show that integers in a trie implementation can be compressed substantially.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Until now, all evaluations of Arabic parsing\u00e2\u0080\u0094including the experiments in the previous section\u00e2\u0080\u0094have assumed gold segmentation.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"2 61.7 64.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Further, the probing hash table does only one random lookup per query, explaining why it is faster on large data.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"If \u00e2\u0080\u009cgun\u00e2\u0080\u009d and \u00e2\u0080\u009crevolver\u00e2\u0080\u009d refer to the same object, then it should also be acceptable to say that Fred was \u00e2\u0080\u009ckilled with a gun\u00e2\u0080\u009d and that the burglar \u00e2\u0080\u009cfireda revolver\u00e2\u0080\u009d.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Performance typically stabilizes across languages after only a few number of iterations.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Cur In order to ground such approaches in linguistic observation and description, a multi-level anno 10 For an exposition of the idea as applied to the task of text planning, see (Chiarcos, Stede 2004).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The second term is a regularizer and encourages all type marginals to be uniform to the extent that is allowed by the first two terms (cf. maximum entropy principle).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In addition, this formulation results in a dramatic reduction in the number of model parameters thereby, enabling unusually rapid training.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The key point is that the second constraint can be remarkably powerful in reducing the complexity of the learning problem.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The method uses a &quot;soft&quot; measure of the agreement between two classifiers as an objective function; we described an algorithm which directly optimizes this function.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"This approach makes the training objective more complex by adding linear constraints proportional to the number of word types, which is rather prohibitive.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J) and the lexicon probability p(fbi jei) are obtained by relative frequency estimates from the Viterbi alignment path after the final training iteration.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Foreign names are usually transliterated using hanzi whose sequential pronunciation mimics the source language pronunciation of the name.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The development set contained 44088 constituents in 2416 sentences and the test set contained 30691 constituents in 1699 sentences.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Each extraction pattern represents a linguistic expression and a syntactic position indicating where a role filler can be found.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Although the best published results for the Collins parser is 80% UAS (Collins, 1999), this parser reaches 82% when trained on the entire training data set, and an adapted version of Charniak\u2019s parser (Charniak, 2000) performs at 84% (Jan Haji\u02c7c, pers. comm.).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This is orthographically represented as 7C.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The paper is organized as follows: Section 2 explains the different layers of annotation that have been produced or are being produced.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Context from the whole document can be important in classifying a named entity.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The Input The set of analyses for a token is thus represented as a lattice in which every arc corresponds to a specific lexeme l, as shown in Figure 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Translation errors are reported in terms of multireference word error rate (mWER) and subjective sentence error rate (SSER).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Human evaluation is one way to distinguish between the two cases.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(Blum and Mitchell 98) describe learning in the following situation: X = X1 X X2 where X1 and X2 correspond to two different &quot;views&quot; of an example.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Words and punctuation that appear in brackets are considered optional.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"6 Conclusions.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Each round is composed of two stages; each stage updates one of the classifiers while keeping the other classifier fixed.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"9 66.4 47.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This is not ideal for some applications, however.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This analyzer setting is similar to that of (Cohen and Smith, 2007), and models using it are denoted nohsp, Parser and Grammar We used BitPar (Schmid, 2004), an efficient general purpose parser,10 together with various treebank grammars to parse the input sentences and propose compatible morphological segmentation and syntactic analysis.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The Expectation Maximization (EM) algorithm (Dempster, Laird and Rubin 77) is a common approach for unsupervised training; in this section we describe its application to the named entity problem.","label":0},{"source_text":"There is no global pruning.","target_text":"The translation scores for the hypotheses generated with different threshold values t0 are compared to the translation scores obtained with a conservatively large threshold t0 = 10:0 . For each test series, we count the number of sentences whose score is worse than the corresponding score of the test series with the conservatively large threshold t0 = 10:0, and this number is reported as the number of search errors.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"On several languages, we report performance exceeding that of state-of-the art systems.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"In focus is in particular the correlation with rhetorical structure, i.e., the question whether specific rhetorical relations \u00e2\u0080\u0094 or groups of relations in particular configurations \u00e2\u0080\u0094 are signalled by speakers with prosodic means.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"This is the parse that is closest to the centroid of the observed parses under the similarity metric.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"It is difficult when IN and OUT are dissimilar, as they are in the cases we study.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Only 2 link in the CC- domain (buy-purchase, acquire-acquisition) and 2 links (trader-dealer and head-chief) in the PC- domain are found in the same synset of Word- Net 2.1 (http:\/\/wordnet.princeton.edu\/).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This section describes AdaBoost, which is the basis for the CoBoost algorithm.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In this paper, Section 2 begins by explaining how contextual role knowledge is represented and learned.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"When the connective is an adverbial, there is much less clarity as to the range of the spans.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As can be seen, GR and this \"pared-down\" statistical method perform quite similarly, though the statistical method is still slightly better.16 AG clearly performs much less like humans than these methods, whereas the full statistical algorithm, including morphological derivatives and names, performs most closely to humans among the automatic methods.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Space- or punctuation-delimited * 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In the CC-domain, there are 32 sets of phrases which contain more than 2 phrases.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"When SRILM estimates a model, it sometimes removes n-grams but not n + 1-grams that extend it to the left.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"M(wi) = Li).","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The choice of the genre commentary resulted from the fact that an investigation of rhetorical structure, its interaction with other aspects of discourse structure, and the prospects for its automatic derivation are the key motivations for building up the corpus.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We have presented a method for unsupervised part- of-speech tagging that considers a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Keywords with more than one word In the evaluation, we explained that \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d are considered paraphrases.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The difference between the featureless model (+PRIOR) and our full model (+FEATS) is 13.6% and 7.7% average error reduction on best and median settings respectively.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"When this feature type was included, CoBoost chose this default feature at an early iteration, thereby giving non-abstaining pseudo-labels for all examples, with eventual convergence to the two classifiers agreeing by assigning the same label to almost all examples.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"08 84.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Two issues distinguish the various proposals.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The ATB annotation guidelines specify that proper nouns should be specified with a flat NP (a).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Reflexive pronouns with only 1 NP in scope..","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"08 84.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"0 57.3 51.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We define the following function: If Zco is small, then it follows that the two classifiers must have a low error rate on the labeled examples, and that they also must give the same label on a large number of unlabeled instances.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Following Sproat and Shih (1990), performance for Chinese segmentation systems is generally reported in terms of the dual measures of precision and recalP It is fairly standard to report precision and recall scores in the mid to high 90% range.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"In contrast to the Bayesian HMM, \u00ce\u00b8t is not drawn from a distribution which has support for each of the n word types.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(a) I f f fi * fi :1 }'l ij 1\u00c2\u00a7: {1M m m s h e n 3 m e 0 shi2 ho u4 wo 3 cai2 ne ng 2 ke4 fu 2 zh e4 ge 4 ku n4 w h a t ti m e I just be abl e ov er co m e thi s C L dif fic 'When will I be able to overcome this difficulty?'","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Most languages that use Roman, Greek, Cyrillic, Armenian, or Semitic scripts, and many that use Indian-derived scripts, mark orthographic word boundaries; however, languages written in a Chinese-derived writ\u00c2\u00ad ing system, including Chinese and Japanese, as well as Indian-derived writing systems of languages like Thai, do not delimit orthographic words.1 Put another way, written Chinese simply lacks orthographic words.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Instead, the designs of the various annotation layers and the actual annotation work are results of a series of diploma theses, of students\u00e2\u0080\u0099 work in course projects, and to some extent of paid assistentships.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Variants of alif are inconsistently used in Arabic texts.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We now describe the CoBoost algorithm for the named entity problem.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Syntactic decoders, such as cdec (Dyer et al., 2010), build state from null context then store it in the hypergraph node for later extension.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Besides information structure, the second main goal is to enhance current models of rhetorical structure.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"These conaUses lossy compression. bThe 8-bit quantized variant returned incorrect probabilities as explained in Section 3.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"It is not immediately obvious how to formulate an equivalent to equation (1) for an adapted TM, because there is no well-defined objective for learning TMs from parallel corpora.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This representation gives ir, an appropriate morphological decomposition, pre\u00c2\u00ad serving information that would be lost by simply listing ir, as an unanalyzed form.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Analysis of the data revealed that the contextual role knowledge is especially helpful for resolving pronouns because, in general, they are semantically weaker than definite NPs.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In the terrorism domain, 1600 texts were used for training and the 40 test docu X \u00e2\u0088\u00a9Y =\u00e2\u0088\u0085 All sets of hypotheses (and their corresponding belief values) in the current model are crossed with the sets of hypotheses (and belief values) provided by the new evidence.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"3.1 Gross Statistics.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This significantly underperforms log-linear combination.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"To set \u03b2, we used the same criterion as for \u03b1, over a dev corpus: The MAP combination was used for TM probabilities only, in part due to a technical difficulty in formulating coherent counts when using standard LM smoothing techniques (Kneser and Ney, 1995).3 Motivated by information retrieval, a number of approaches choose \u201crelevant\u201d sentence pairs from OUT by matching individual source sentences from IN (Hildebrand et al., 2005; L\u00a8u et al., 2007), or individual target hypotheses (Zhao et al., 2004).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"See Section 5.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The PCFG was trained from the same sections of the Penn Treebank as the other three parsers.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The final model tions.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"It is. based on the traditional character set rather than the simplified character set used in Singapore and Mainland China.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Table 9: Dev set results for sentences of length \u00e2\u0089\u00a4 70.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","label":1},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Using structural information As was explained in the results section, we extracted examples like \u00e2\u0080\u009cSmith estimates Lotus\u00e2\u0080\u009d, from a sentence like \u00e2\u0080\u009cMr.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"However, using the top-level semantic classes of WordNet proved to be problematic because the class distinctions are too coarse.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We used a standard one-pass phrase-based system (Koehn et al., 2003), with the following features: relative-frequency TM probabilities in both directions; a 4-gram LM with Kneser-Ney smoothing; word-displacement distortion model; and word count.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Our original hope in combining these parsers is that their errors are independently distributed.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The cost is computed as follows, where N is the corpus size and f is the frequency: (1) Besides actual words from the base dictionary, the lexicon contains all hanzi in the Big 5 Chinese code\/ with their pronunciation(s), plus entries for other characters that can be found in Chinese text, such as Roman letters, numerals, and special symbols.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"While we have minimized forward-looking state in Section 4.1, machine translation systems could also benefit by minimizing backward-looking state.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Compared with the widely- SRILM, our is 2.4 times as fast while using 57% of the mem- The structure is a trie with bit-level packing, sorted records, interpolation search, and optional quantization aimed lower memory consumption. simultaneously uses less memory than the smallest lossless baseline and less CPU than the baseline.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The frequency of the Company \u00e2\u0080\u0093 Company domain ranks 11th with 35,567 examples.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Recently, lattices have been used successfully in the parsing of Hebrew (Tsarfaty, 2006; Cohen and Smith, 2007), a Semitic language with similar properties to Arabic.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Such constraints are derived from training data, expressing some relationship between features and outcome.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We are especially grateful to Taylor Berg- Kirkpatrick for running additional experiments.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"For seven out of eight languages a threshold of 0.2 gave the best results for our final model, which indicates that for languages without any validation set, r = 0.2 can be used.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"With the same Corporate- Suffix-List and Person-Prefix-List used in local features, for a token seen elsewhere in the same document with one of these suffixes (or prefixes), another feature Other-CS (or Other-PP) is set to 1.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"5 \u00e2\u0080\u0098Underspecified Rhetorical Markup Language\u00e2\u0080\u0099 6 This confirms the figure given by (Schauer, Hahn.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Figure 5 shows how this model is implemented as part of the dictionary WFST.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The second algorithm builds on a boosting algorithm called AdaBoost (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"As is standard, we use a fixed constant K for the number of tagging states.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The Expectation Maximization (EM) algorithm (Dempster, Laird and Rubin 77) is a common approach for unsupervised training; in this section we describe its application to the named entity problem.","label":0},{"source_text":"There is no global pruning.","target_text":"The perplexity for the trigram language model used is 26:5.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"\u00e2\u0080\u00a2 Anaphoric links: the annotator is asked to specify whether the anaphor is a repetition, partial repetition, pronoun, epithet (e.g., Andy Warhol \u00e2\u0080\u0093 the PopArt artist), or is-a (e.g., Andy Warhol was often hunted by photographers.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Saving state allows our code to walk the data structure exactly once per query.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The following two error criteria are used in our experiments: mWER: multi-reference WER: We use the Levenshtein distance between the automatic translation and several reference translations as a measure of the translation errors.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We make the assumption that for each example, both xi,. and x2,2 alone are sufficient to determine the label yi.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Note that Chang, Chen, and Chen (1991), in addition to word-frequency information, include a constraint-satisfication model, so their method is really a hybrid approach.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":" We evaluate the system's performance by comparing its segmentation 'Tudgments\" with the judgments of a pool of human segmenters, and the system is shown to perform quite well.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"On several languages, we report performance exceeding that of state-of-the art systems.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Here, we process only full-form words within the translation procedure.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Proper-Name Identification.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Another question that remains unanswered is to what extent the linguistic information he considers can be handled-or at least approximated-by finite-state language models, and therefore could be directly interfaced with the segmentation model that we have presented in this paper.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The terrorism examples reflect fairly obvious relationships: people who are murdered are killed; agents that \u00e2\u0080\u009creport\u00e2\u0080\u009d things also \u00e2\u0080\u009cadd\u00e2\u0080\u009d and \u00e2\u0080\u009cstate\u00e2\u0080\u009d things; crimes that are \u00e2\u0080\u009cperpetrated\u00e2\u0080\u009d are often later \u00e2\u0080\u009ccondemned\u00e2\u0080\u009d.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The type-level posterior term can be computed according to, P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2) \u00e2\u0088\u009d Note that each round of sampling Ti variables takes time proportional to the size of the corpus, as with the standard token-level HMM.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"7 www.cis.upenn.edu\/\u00e2\u0088\u00bcpdtb\/ 8 www.eml-research.de\/english\/Research\/NLP\/ Downloads had to buy a new car.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"As can be seen in the example, the first two phrases have a different order of NE names from the last two, so we can determine that the last two phrases represent a reversed relation.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"CFG's, TAG's, MCTAG's and HG's are all members of this class since they satisfy these restrictions.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"These are shown, with their associated costs, as follows: ABj nc 4.0 AB C\/jj 6.0 CD \/vb 5.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"2.1 Part-of-speech tags.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Within the RST \u00e2\u0080\u009cuser community\u00e2\u0080\u009d there has also been discussion whether two levels of discourse structure should not be systematically distinguished (intentional versus informational).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If is a number string (such as one, two, etc), then the feature NumberString is set to 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Additionally, it works about 3 times as fast as the IBM style search.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We thank United Informatics for providing us with our corpus of Chinese text, and BDC for the 'Behavior ChineseEnglish Electronic Dictionary.'","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The dependency graph in Figure 1 satisfies all the defining conditions above, but it fails to satisfy the condition ofprojectivity (Kahane et al., 1998): The arc connecting the head jedna (one) to the dependent Z (out-of) spans the token je (is), which is not dominated by jedna.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As we have seen, the lexicon of basic words and stems is represented as a WFST; most arcs in this WFST represent mappings between hanzi and pronunciations, and are costless.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Many statistical or machine-learning approaches for natural language problems require a relatively large amount of supervision, in the form of labeled training examples.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Consequently, we implemented our own annotation tool ConAno in Java (Stede, Heintze 2004), which provides specifically the functionality needed for our purpose.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"6.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"At the same time, the n-gram error rate is sensitive to samples with extreme n-gram counts.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Before describing the unsupervised case we first describe the supervised version of the algorithm: Input to the learning algorithm: n labeled examples of the form (xi, y\u201e). y, is the label of the ith example (given that there are k possible labels, y, is a member of y = {1 ... 0). xi is a set of mi features {x,1, Xi2 .","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We measured recall (Rec), precision (Pr), and the F-measure (F) with recall and precision equally weighted.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"It uses a maximum entropy framework and classifies each word given its features.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The general idea for the knowledge- based part is to have the system use as much information as it can find at its disposal to produce a target representation as specific as possible and as underspecified as necessary.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Word type N % Dic tion ary entr ies 2 , 5 4 3 9 7 . 4 7 Mor pho logi call y deri ved wor ds 3 0 . 1 1 Fore ign tran slite rati ons 9 0 . 3 4 Per son al na mes 5 4 2 . 0 7 cases.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Memory usage is the same as with binary search and lower than with set.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":1},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Roughly speaking, previous work can be divided into three categories, namely purely statistical approaches, purely lexi\u00c2\u00ad cal rule-based approaches, and approaches that combine lexical information with sta\u00c2\u00ad tistical information.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Our work is closest to that of Yarowsky and Ngai (2001), but differs in two important ways.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"This is the form of recursive levels in iDafa constructs.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The PCFG was trained from the same sections of the Penn Treebank as the other three parsers.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Indeed, as we shall show in Section 5, even human judges differ when presented with the task of segmenting a text into words, so a definition of the criteria used to determine that a given segmentation is correct is crucial before one can interpret such measures.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The human evaluators were a non-native, fluent Arabic speaker (the first author) for the ATB and a native English speaker for the WSJ.7 Table 5 shows type- and token-level error rates for each corpus.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"ICOC and CSPP contributed the greatest im provements.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The quasi-monotone search performs best in terms of both error rates mWER and SSER.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"AdaBoost.MH can be applied to the problem using these pseudolabels in place of supervised examples.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"First, we parsed the training corpus, collected all the noun phrases, and looked up each head noun in WordNet (Miller, 1990).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"For example, in TAG's a derived auxiliary tree spans two substrings (to the left and right of the foot node), and the adjunction operation inserts another substring (spanned by the subtree under the node where adjunction takes place) between them (see Figure 3).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We observe similar trends when using another measure \u00e2\u0080\u0093 type-level accuracy (defined as the fraction of words correctly assigned their majority tag), according to which La ng ua ge M etr ic B K 10 E M B K 10 L B F G S G 10 F EA T S B es t F EA T S M ed ia n E ng lis h 1 1 m 1 4 8 . 3 6 8 . 1 5 6 . 0 7 5 . 5 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 0 . 9 6 6 . 4 4 7 . 8 6 6 . 4 D an is h 1 1 m 1 4 2 . 3 6 6 . 7 4 2 . 6 5 8 . 0 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 2 . 1 6 1 . 2 4 3 . 2 6 0 . 7 D ut ch 1 1 m 1 5 3 . 7 6 7 . 0 5 5 . 1 6 4 . 7 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 6 . 4 6 9 . 0 5 1 . 5 6 7 . 3 Po rtu gu es e 1 1 m 1 5 0 . 8 7 5 . 3 4 3 . 2 7 4 . 8 44 .5 69 .2 6 4 . 1 7 4 . 5 5 6 . 5 7 0 . 1 S pa ni sh 1 1 m 1 \u00e2\u0080\u0093 \u00e2\u0080\u0093 4 0 . 6 7 3 . 2 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 8 . 3 6 8 . 9 5 0 . 0 5 7 . 2 Table 4: Comparison of our method (FEATS) to state-of-the-art methods.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"In addition, as the CRF and PCFG look at similar sorts of information from within two inherently different models, they are far from independent and optimizing their product is meaningless.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We have shown that the maximum entropy framework is able to use global information directly.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"KenLM: Faster and Smaller Language Model Queries","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We place certain restrictions on the composition operations of LCFRS's, restrictions that are shared by the composition operations of the constrained grammatical systems that we have considered.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Two common cases are the attribu tive adjective and the process nominal _; maSdar, which can have a verbal reading.4 At tributive adjectives are hard because they are or- thographically identical to nominals; they are inflected for gender, number, case, and definiteness.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In (Bikel et al., 1997) and (Bikel et al., 1999), performance was plotted against training data size to show how performance improves with training data size.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"With the additional assumptions, inspired by Rounds (1985), we can show that members of this class can be recognized in polynomial time.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In Eq.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Therefore, for n-gram wn1 , all leftward extensions wn0 are an adjacent block in the n + 1-gram array.","label":0},{"source_text":"There is no global pruning.","target_text":"However, dynamic programming can be used to find the shortest tour in exponential time, namely in O(n22n), using the algorithm by Held and Karp.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Our experiments all concern the analytical annotation, and the first experiment is based only on the training part.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Before we turn to the evaluation, however, we need to introduce the data-driven dependency parser used in the latter experiments.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The machine learning community has been in a similar situation and has studied the combination of multiple classifiers (Wolpert, 1992; Heath et al., 1996).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Sometimes, however, these beliefs can be contradictory.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The final score is obtained from: max e;e0 j2fJ\u00f4\u0080\u0080\u0080L;;Jg p($je; e0) Qe0 (e; I; f1; ; Jg; j); where p($je; e0) denotes the trigram language model, which predicts the sentence boundary $ at the end of the target sentence.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Evaluation of links A link between two sets is considered correct if the majority of phrases in both sets have the same meaning, i.e. if the link indicates paraphrase.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Otherwise, the scope of the search problem shrinks recursively: if A[pivot] < k then this becomes the new lower bound: l +\u2014 pivot; if A[pivot] > k then u +\u2014 pivot.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"A compatible view is presented by Charniak et al. (1996) who consider the kind of probabilities a generative parser should get from a PoS tagger, and concludes that these should be P(w|t) \u201cand nothing fancier\u201d.3 In our setting, therefore, the Lattice is not used to induce a probability distribution on a linear context, but rather, it is used as a common-denominator of state-indexation of all segmentations possibilities of a surface form.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"evaluated to account for the same fraction of the data.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We concentrate on those sets.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Each extraction pattern represents a linguistic expression and a syntactic position indicating where a role filler can be found.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This model admits a simple Gibbs sampling algorithm where the number of latent variables is proportional to the number of word types, rather than the size of a corpus as for a standard HMM sampler (Johnson, 2007).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Mi(c) is a binary function returning t when parser i (from among the k parsers) suggests constituent c should be in the parse.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"On the English side, however, the vertices (denoted by Ve) correspond to word types.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This paper proposes a simple and effective tagging method that directly models tag sparsity and other distributional properties of valid POS tag assignments.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In many cases these failures in recall would be fixed by having better estimates of the actual prob\u00c2\u00ad abilities of single-hanzi words, since our estimates are often inflated.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"It is difficult to directly compare the Matsoukas et al results with ours, since our out-of-domain corpus is homogeneous; given heterogeneous training data, however, it would be trivial to include Matsoukas-style identity features in our instance-weighting model.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Next, we represent the input sentence as an unweighted finite-state acceptor (FSA) I over H. Let us assume the existence of a function Id, which takes as input an FSA A, and produces as output a transducer that maps all and only the strings of symbols accepted by A to themselves (Kaplan and Kay 1994).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Precision.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The toplevel weights are trained to maximize a metric such as BLEU on a small development set of approximately 1000 sentence pairs.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"), which precludes a single universal approach to adaptation.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"We apply a beam search concept as in speech recognition.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We further thank Dr. J.-S.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Second, we show that although the Penn Arabic Treebank is similar to other tree- banks in gross statistical terms, annotation consistency remains problematic.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We have argued that the proposed method performs well.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Smith estimates Lotus will make a profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d, our system extracts \u00e2\u0080\u009cSmith esti mates Lotus\u00e2\u0080\u009d as an instance.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"It is difficult to directly compare the Matsoukas et al results with ours, since our out-of-domain corpus is homogeneous; given heterogeneous training data, however, it would be trivial to include Matsoukas-style identity features in our instance-weighting model.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Only IRSTLM does not support threading.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"On the other hand, we are interested in the application of rhetorical analysis or \u00e2\u0080\u0098discourse parsing\u00e2\u0080\u0099 (3.2 and 3.3), in text generation (3.4), and in exploiting the corpus for the development of improved models of discourse structure (3.5).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Chinese word segmentation can be viewed as a stochastic transduction problem.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We trained this model by optimizing the following objective function: Note that this involves marginalizing out all possible state configurations z for a sentence x, resulting in a non-convex objective.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For each extension a new position is added to the coverage set.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Since MUC6, BBN' s Hidden Markov Model (HMM) based IdentiFinder (Bikel et al., 1997) has achieved remarkably good performance.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The corpora for both settings are summarized in table 1.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Note that Zt is a normalization constant that ensures the distribution Dt+i sums to 1; it is a function of the weak hypothesis ht and the weight for that hypothesis at chosen at the tth round.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(2009b) evaluated the Bikel parser using the same ATB split, but only reported dev set results with gold POS tags for sentences of length \u00e2\u0089\u00a4 40.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"For each source word f, the list of its possible translations e is sorted according to p(fje) puni(e), where puni(e) is the unigram probability of the English word e. It is su\u00c3\u0086cient to consider only the best 50 words.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"But Rehbein and van Genabith (2007) showed that Evalb should not be used as an indication of real difference\u00e2\u0080\u0094 or similarity\u00e2\u0080\u0094between treebanks.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Comparison with state-of-the-art taggers For comparison we consider two unsupervised tag- gers: the HMM with log-linear features of Berg- Kirkpatrick et al.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"gaolgaolxing4xing4 'happily' In the particular form of A-not-A reduplication illustrated in (3a), the first syllable of the verb is copied, and the negative markerbu4 'not' is inserted between the copy and the full verb.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"\u00e2\u0080\u009cThe gun\u00e2\u0080\u009d will be extracted by the caseframe \u00e2\u0080\u009cfired <patient>\u00e2\u0080\u009d.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We use a squared loss to penalize neighboring vertices that have different label distributions: kqi \u2212 qjk2 = Ey(qi(y) \u2212 qj(y))2, and additionally regularize the label distributions towards the uniform distribution U over all possible labels Y.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Recent work has made significant progress on unsupervised POS tagging (Me\u00c2\u00b4rialdo, 1994; Smith and Eisner, 2005; Haghighi and Klein, 2006; Johnson,2007; Goldwater and Griffiths, 2007; Gao and John son, 2008; Ravi and Knight, 2009).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Some approaches depend upon some form of con\u00c2\u00ad straint satisfaction based on syntactic or semantic features (e.g., Yeh and Lee [1991], which uses a unification-based approach).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Top 5 Bot to m 5 Go ld NN P NN JJ CD NN S RB S PD T # \u00e2\u0080\u009d , 1T W CD W RB NN S VB N NN PR P$ W DT : MD . +P RI OR CD JJ NN S WP $ NN RR B- , $ \u00e2\u0080\u009d . +F EA TS JJ NN S CD NN P UH , PR P$ # . \u00e2\u0080\u009c Table 5: Type-level English POS Tag Ranking: We list the top 5 and bottom 5 POS tags in the lexicon and the predictions of our models under the best hyperparameter setting.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We report micro-averaged (whole corpus) and macro-averaged (per sentence) scores along add a constraint on the removal of punctuation, which has a single tag (PUNC) in the ATB.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"This process is repeated 5 times by rotating the data appropriately.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The P (T |\u00cf\u0088) distribution, in English for instance, should have very low mass for the DT (determiner) tag, since determiners are a very small portion of the vocabulary.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This is consistent with the nature of these two settings: log-linear combination, which effectively takes the intersection of IN and OUT, does relatively better on NIST, where the domains are broader and closer together.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"For example, from the sentence \u00e2\u0080\u009cMr.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This variant is tested in the experiments below.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"In Section 5, we try to compare results of MENE, IdentiFinder, and MENERGI.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Jiang and Zhai (2007) suggest the following derivation, making use of the true OUT distribution po(s, t): where each fi(s, t) is a feature intended to charac- !0\u02c6 = argmax pf(s, t) log p\u03b8(s|t) (8) terize the usefulness of (s, t), weighted by Ai. \u03b8 s,t pf(s, t)po(s, t) log p\u03b8(s|t) The mixing parameters and feature weights (col- != argmax po (s, t) lectively 0) are optimized simultaneously using dev- \u03b8 s,t pf(s, t)co(s, t) log p\u03b8(s|t), set maximum likelihood as before: !\ufffdargmax po (s, t) ! \u03b8 s,t \ufffd\u02c6 = argmax \u02dcp(s, t) log p(s|t; 0).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"5 67.3 55.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Other approaches encode sparsity as a soft constraint.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Our empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Different sentence structure and rich target language morphology are two reasons for this.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Any opinions, findings, conclusions, or recommendations expressed in this paper are those of the authors, and do not necessarily reflect the views of the funding organizations.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We incur some additional memory cost due to storing state in each hypothesis, though this is minimal compared with the size of the model itself.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Our second point of comparison is with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"6 Results and Analysis.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"f;g denotes the empty set, where no source sentence position is covered.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"For each domain, phrases which contain the same keyword are gathered to build a set of phrases (Step 3).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"While our method also enforces a singe tag per word constraint, it leverages the transition distribution encoded in an HMM, thereby benefiting from a richer representation of context.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Thus we are interested not in extraction, but actual generation from representations that may be developed to different degrees of granularity.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(Other classes handled by the current system are discussed in Section 5.)","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Recently, combination techniques have been investigated for part of speech tagging with positive results (van Halteren et al., 1998; Brill and Wu, 1998).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In the second part of the experiment, we applied the inverse transformation based on breadth-first search under the three different encoding schemes.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"buy - acquire (5) buy - agree (2) buy - purchase (5) buy - acquisition (7) buy - pay (2)* buy - buyout (3) buy - bid (2) acquire - purchase (2) acquire - acquisition (2) acquire - pay (2)* purchase - acquisition (4) purchase - stake (2)* acquisition - stake (2)* unit - subsidiary (2) unit - parent (5) It is clear that these links form two clusters which are mostly correct.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The techniques we develop can be extended in a relatively straightforward manner to the more general case when OUT consists of multiple sub-domains.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Clearly this poses a number of research challenges, though, such as the applicability of tag sets across different languages.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"As the two NE categories are the same, we can\u00e2\u0080\u0099t differentiate phrases with different orders of par ticipants \u00e2\u0080\u0093 whether the buying company or the to-be-bought company comes first.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Thus, the tree sets generated by HG's are similar to those of CFG's, with each node annotated by the operation (concatenation or wrapping) used to combine the headed strings derived by the daughters of Tree Adjoining Grammars, a tree rewriting formalism, was introduced by Joshi, Levy and Takahashi (1975) and Joshi (1983\/85).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Here, we process only full-form words within the translation procedure.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"In Semitic languages the situation is very different.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For example, if X and Y are coreferent, then both X and Y are considered to co-occur with the caseframe that extracts X as well as the caseframe that extracts Y. We will refer to the set of nouns that co-occur with a caseframe as the lexical expectations of the case- frame.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"RandLM is the clear winner in RAM utilization, but is also slower and lower quality.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Despite these limitations, a purely finite-state approach to Chinese word segmentation enjoys a number of strong advantages.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The Potsdam Commentary Corpus","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Table 3 contains the results for evaluating our systems on the test set (section 22).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This has solutions: where pI(s|t) is derived from the IN corpus using relative-frequency estimates, and po(s|t) is an instance-weighted model derived from the OUT corpus.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"(f1; ;mg n fl1g ; l) 3 (f1; ;mg n fl; l1; l2g ; l0) !","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"For example, the phrase \u00e2\u0080\u009c's New York-based trust unit,\u00e2\u0080\u009d is not a paraphrase of the other phrases in the \u00e2\u0080\u009cunit\u00e2\u0080\u009d set.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"In this paper we study the problem of using a parallel corpus from a background domain (OUT) to improve performance on a target domain (IN) for which a smaller amount of parallel training material\u2014though adequate for reasonable performance\u2014is also available.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"So the success of the algorithm may well be due to its success in maximizing the number of unlabeled examples on which the two decision lists agree.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A greedy algorithm (or maximum-matching algorithm), GR: proceed through the sentence, taking the longest match with a dictionary entry at each point.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Approaches differ in the algorithms used for scoring and selecting the best path, as well as in the amount of contextual information used in the scoring process.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"). context=x The context for the entity.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In this paper, we show that the maximum entropy framework is able to make use of global information directly, and achieves performance that is comparable to the best previous machine learning-based NERs on MUC6 and MUC7 test data.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We attain these results using several optimizations: hashing, custom lookup tables, bit-level packing, and state for left-to-right query patterns.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Finally, model GTv = 2 includes parent annotation on top of the various state-splits, as is done also in (Tsarfaty and Sima\u2019an, 2007; Cohen and Smith, 2007).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"10 Chinese speakers may object to this form, since the suffix f, menD (PL) is usually restricted to.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Yarowsky-cautious does not separate the spelling and contextual features, but does have a limit on the number of rules added at each stage.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"As already pointed out in Section 2.4, current theories diverge not only on the number and definition of relations but also on apects of structure, i.e., whether a tree is sufficient as a representational device or general graphs are required (and if so, whether any restrictions can be placed on these graph\u00e2\u0080\u0099s structures \u00e2\u0080\u0094 cf.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"It can be shown that this objective is convex in q.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"so that 'door' would be and in this case the hanzi 7C, does not represent a syllable.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"A different notion of information structure, is used in work such as that of (?), who tried to characterize felicitous constituent ordering (theme choice, in particular) that leads to texts presenting information in a natural, \u00e2\u0080\u009cflowing\u00e2\u0080\u009d way rather than with abrupt shifts of attention.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For graph propagation, the hyperparameter v was set to 2 x 10\u22126 and was not tuned.","label":0},{"source_text":"There is no global pruning.","target_text":"For the translation experiments, Eq. 2 is recursively evaluated.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, if X and Y are coreferent, then both X and Y are considered to co-occur with the caseframe that extracts X as well as the caseframe that extracts Y. We will refer to the set of nouns that co-occur with a caseframe as the lexical expectations of the case- frame.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"A detailed description of the search procedure used is given in this patent.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For a sequence of hanzi that is a possible name, we wish to assign a probability to that sequence qua name.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"com \u00c2\u00a7Cambridge, UK Email: nc201@eng.cam.ac.uk \u00c2\u00a9 1996 Association for Computational Linguistics (a) B ) ( , : & ; ? ' H o w d o y o u s a y o c t o p u s i n J a p a n e s e ? ' (b) P l a u s i b l e S e g m e n t a t i o n I B X I I 1 : & I 0 0 r i 4 w e n 2 z h a n g l y u 2 z e n 3 m e 0 s h u o l ' J a p a n e s e ' ' o c t o p u s ' ' h o w ' ' s a y ' (c) Figure 1 I m p l a u s i b l e S e g m e n t a t i o n [\u00c2\u00a7] lxI 1:&I ri4 wen2 zhangl yu2zen3 me0 shuol 'Japan' 'essay' 'fish' 'how' 'say' A Chinese sentence in (a) illustrating the lack of word boundaries.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We make the assumption that for each example, both xi,. and x2,2 alone are sufficient to determine the label yi.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In the case of, the most common usage is as an adverb with the pronunciation jiangl, so that variant is assigned the estimated cost of 5.98, and a high cost is assigned to nominal usage with the pronunciation jiang4.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"For example, the two NEs \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d and \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d have the following contexts.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We extend the Stanford parser to accept pre-generated lattices, where each word is represented as a finite state automaton.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In the numerator, however, the counts of ni1s are quite irregular, in\u00c2\u00ad cluding several zeros (e.g., RAT, none of whose members were seen).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"A different knowledge source, called CFSemCFSem, compares the semantic expectations of the caseframe that extracts the anaphor with the semantic expectations of the caseframe that extracts the candidate.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The sentence length probability p(JjI) is omitted without any loss in performance.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In this paper we have argued that Chinese word segmentation can be modeled ef\u00c2\u00ad fectively using weighted finite-state transducers.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Using cn to denote the number of n-grams, total memory consumption of TRIE, in bits, is plus quantization tables, if used.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Our evaluation includes both weighted and un- weighted lattices.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The overall performance of the LTG system was outstanding, but the system consists of a sequence of many hand-coded rules and machine-learning modules.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"5 68.1 34.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"(3)In sentence (1), McCann can be a person or an orga nization.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This is orthographically represented as 7C.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"If any of the tokens from to is in Person-Prefix- List, then another feature Person-Prefix is set to 1.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The effect of UNIQ is very small on both data sets.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"This approach leads to a search procedure with complexity O(E3 J4).","label":0},{"source_text":"Here we present two algorithms.","target_text":"For example, take ..., says Maury Cooper, a vice president at S.&P.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We evaluated the results based on two metrics.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The relativizer f(\u201cthat\u201d) for example, may attach to an arbitrarily long relative clause that goes beyond token boundaries.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Only one of the first n positions which are not already aligned in a partial hypothesis may be chosen, where n is set to 4.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We elected run Moses single-threaded to minimize the impact of RandLM\u2019s cache on memory use.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The algorithm builds two classifiers in parallel from labeled and unlabeled data.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The segmentation chosen is the best path through the WFST, shown in (d).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Matsoukas et al (2009) generalize it by learning weights on sentence pairs that are used when estimating relative-frequency phrase-pair probabilities.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"). context=x The context for the entity.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Similarly Figures 1 and 2 show how the isolated constituent precision varies by sentence length and the size of the span of the hypothesized constituent.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Case and Zone: If the token starts with a capital letter (initCaps), then an additional feature (init- Caps, zone) is set to 1.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"). context=x The context for the entity.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"This algorithm can be applied to statistical machine translation.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Floating point values may be stored in the trie exactly, using 31 bits for non-positive log probability and 32 bits for backoff5.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Only one of the first n positions which are not already aligned in a partial hypothesis may be chosen, where n is set to 4.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The code is opensource, has minimal dependencies, and offers both C++ and Java interfaces for integration.","label":1},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(Blum and Mitchell 98) offer a promising formulation of redundancy, also prove some results about how the use of unlabeled examples can help classification, and suggest an objective function when training with unlabeled examples.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Recent work by Finkel and Manning (2009) which re-casts Daum\u00b4e\u2019s approach in a hierarchical MAP framework may be applicable to this problem.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The weak hypothesis chosen was then restricted to be a predictor in favor of this label.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"While there might be some controversy about the exact definition of such a tagset, these 12 categories cover the most frequent part-of-speech and exist in one form or another in all of the languages that we studied.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The system described in this paper is similar to the MENE system of (Borthwick, 1999).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Also surprising is the low test set OOV rate given the possibility of morphological variation in Arabic.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"So, who won the competition?","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Extract NE pair instances with contexts From the four years of newspaper corpus, we extracted 1.9 million pairs of NE instances.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"However, when we pre- tag the input\u00e2\u0080\u0094as is recommended for English\u00e2\u0080\u0094 we notice a 0.57% F1 improvement.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"12 One class of full personal names that this characterization does not cover are married women's names.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"nan2gual 'pumpkin.'","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Another approach to finding paraphrases is to find phrases which take similar subjects and objects in large corpora by using mutual information of word distribution [Lin and Pantel 01].","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"So, we set a threshold that at least two examples are required to build a link.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"It was also proposed to allow annotators to skip sentences that they are unable to judge.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The state-to-tag mapping is obtained from the best hyperparameter setting for 11 mapping shown in Table 3.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Our baseline for all sentence lengths is 5.23% F1 higher than the best previous result.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Word type N % Dic tion ary entr ies 2 , 5 4 3 9 7 . 4 7 Mor pho logi call y deri ved wor ds 3 0 . 1 1 Fore ign tran slite rati ons 9 0 . 3 4 Per son al na mes 5 4 2 . 0 7 cases.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The Grammar Our parser looks for the most likely tree spanning a single path through the lattice of which the yield is a sequence of lexemes.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"(f1; ;mg ; l) 2 (f1; ;mg n fl; l1g ; l0) !","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"If 7 is an elementary tree, the derivation tree consists of a single node labeled 7.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The current work treats both segmental and super-segmental phenomena, yet we note that there may be more adequate ways to treat supersegmental phenomena assuming Word-Based morphology as we explore in (Tsarfaty and Goldberg, 2008).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"This design leads to a significant reduction in the computational complexity of training and inference.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"This property reflects an important aspect of the underlying linguistic theory associated with the formalism.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Since the segmentation corresponds to the sequence of words that has the lowest summed unigram cost, the segmenter under discussion here is a zeroth-order model.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Tsarfaty and Sima\u2019an (2007) have reported state-of-the-art results on Hebrew unlexicalized parsing (74.41%) albeit assuming oracle morphological segmentation.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The maximum entropy framework estimates probabilities based on the principle of making as few assumptions as possible, other than the constraints imposed.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We generate these caseframes automatically by running AutoSlog over the training corpus exhaustively so that it literally generates a pattern to extract every noun phrase in the corpus.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"6.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Our second point of comparison is with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Then we ran binary search to determine the least amount of memory with which it would run.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Using a wide-coverage morphological analyzer based on (Itai et al., 2006) should cater for a better coverage, and incorporating lexical probabilities learned from a big (unannotated) corpus (cf.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"gaolbu4-gaolxing4 (hap-not-happy) 'happy?'","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Rather we believe several methods have to be developed using different heuristics to discover wider variety of paraphrases.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"A graph D = (W, A) is well-formed iff it is acyclic and connected.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The dependency graph in Figure 1 satisfies all the defining conditions above, but it fails to satisfy the condition ofprojectivity (Kahane et al., 1998): The arc connecting the head jedna (one) to the dependent Z (out-of) spans the token je (is), which is not dominated by jedna.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Kollege.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Section 2.1 describes how BABAR generates training examples to use in the learning process.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"There is a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099 of ten commentaries, for which the range of information (except for syntax) has been completed; the remaining data has been annotated to different degrees, as explained below.","label":1},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"3 These are not full case frames in the traditional sense, but they approximate a simple case frame with a single slot.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"In fact, we found that enabling IRSTLM\u2019s cache made it slightly slower, so results in Table 1 use IRSTLM without caching.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"1 53.8 47.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"By using the output of a hand-coded system such as Proteus, MENE can improve its performance, and can even outperform IdentiFinder (Borthwick, 1999).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The 1-bit sign is almost always negative and the 8-bit exponent is not fully used on the range of values, so in practice this corresponds to quantization ranging from 17 to 20 total bits.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Note that because we extracted only high-confidence alignments, many foreign vertices will not be connected to any English vertices.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In the current work morphological analyses and lexical probabilities are derived from a small Treebank, which is by no means the best way to go.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"There are still some open issues to be resolved with the format, but it represents a first step.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"A non-optimal analysis is shown with dotted lines in the bottom frame.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Search CPU time mWER SSER Method [sec] [%] [%] MonS 0:9 42:0 30:5 QmS 10:6 34:4 23:8 IbmS 28:6 38:2 26:2 4.2 Performance Measures.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"pre-processing.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"For our experiments we also report the mean of precision and recall, which we denote by (P + R)I2 and F-measure.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"J:j:l :zhongl :0.0 ;m,Jlong4 :0.0 (mHHaryg9tltHBI) \u00c2\u00a3: _ADV: 5.98 \u00c2\u00a5 :hua2:o.o E :_NC: 4.41 :mln2:o.o mm : guo2 : 0.0 (RopubllcofChlna) .....,.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We incorporate instance weighting into a mixture-model framework, and find that it yields consistent improvements over a wide range of baselines.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"40 75.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"9 50.2 +P RI OR be st me dia n 47.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"On MUC6 data, MENERGI also achieves performance comparable to IdentiFinder when trained on similar amount of training data.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We call this approach parser switching.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models (e.g., (Grosz et al., 1995; Grosz and Sidner, 1998)), syntactic algorithms (e.g., (Hobbs, 1978; Lappin and Le- ass, 1994)), and supervised machine learning systems (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Language models are widely applied in natural language processing, and applications such as machine translation make very frequent queries.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"As noted in Section 4.4, disk cache state is controlled by reading the entire binary file before each test begins.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Thus in a two-hanzi word like lflli?J zhong1guo2 (middle country) 'China' there are two syllables, and at the same time two morphemes.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The dev corpus was taken from the NIST05 evaluation set, augmented with some randomly-selected material reserved from the training set.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"constitute names, since we have only their segmentation, not the actual classification of the segmented words.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"IRSTLM\u2019s quantized variant is the inspiration for our quantized variant.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The P (T |\u00cf\u0088) distribution, in English for instance, should have very low mass for the DT (determiner) tag, since determiners are a very small portion of the vocabulary.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Even without features, but still using the tag prior, our median result is 52.0%, still significantly outperforming Grac\u00c2\u00b8a et al.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This FSA I can be segmented into words by composing Id(I) with D*, to form the WFST shown in Figure 2(c), then selecting the best path through this WFST to produce the WFST in Figure 2(d).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Finally, we find links between sets of phrases, based on the NE instance pair data (for example, different phrases which link \u00e2\u0080\u009cIBM\u00e2\u0080\u009d and \u00e2\u0080\u009cLotus\u00e2\u0080\u009d) (Step 4).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The translation direction is from German to English.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"e0; e are the last two target words, C is a coverage set for the already covered source positions and j is the last position visited.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"This number must be less than or equal to n \u00f4\u0080\u0080\u0080 1.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Two annotators received training with the RST definitions and started the process with a first set of 10 texts, the results of which were intensively discussed and revised.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"There is no global pruning.","target_text":"The quasi-monotone search performs best in terms of both error rates mWER and SSER.","label":0},{"source_text":"There is no global pruning.","target_text":"vierten 12.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Among these are words derived by various productive processes, including: 1.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Having found (spelling, context) pairs in the parsed data, a number of features are extracted.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"3An English sentence with ambiguous PoS assignment can be trivially represented as a lattice similar to our own, where every pair of consecutive nodes correspond to a word, and every possible PoS assignment for this word is a connecting arc.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"When the maSdar lacks a determiner, the constituent as a whole resem bles the ubiquitous annexation construct \u00ef\u00bf\u00bd ?f iDafa.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"each word in the lexicon whether or not each string is actually an instance of the word in question.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The improvement is due to the cost of bit-level reads and avoiding reads that may fall in different virtual memory pages.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Formally, for a lexicon L and segments I \u00e2\u0088\u0088 L, O \u00e2\u0088\u0088\/ L, each word automaton accepts the language I\u00e2\u0088\u0097(O + I)I\u00e2\u0088\u0097.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"But we also need an estimate of the probability for a non-occurring though possible plural form like i\u00c2\u00a5JJ1l.f, nan2gua1-men0 'pumpkins.'","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Each feature concept is akin to a random variable and its occurrence in the text corresponds to a particular instantiation of that random variable.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The subtree under,; is excised from 7, the tree 7' is inserted in its place and the excised subtree is inserted below the foot of y'.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"On several languages, we report performance exceeding that of more complex state-of-the art systems.1","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"For unaligned words, we set the tag to the most frequent tag in the corresponding treebank.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"However, this optimistic search would not visit the entries necessary to store backoff information in the outgoing state.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, if is found in the list of person first names, the feature PersonFirstName is set to 1.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"However, the accuracy is considerably higher than previously reported results for robust non-projective parsing of Czech, with a best performance of 73% UAS (Holan, 2004).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We feel that information from a whole corpus might turn out to be noisy if the documents in the corpus are not of the same genre.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"None of the parsers produce parses with crossing brackets, so none of them votes for both of the assumed constituents.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Final (F): The rest of the sentence is processed monotonically taking account of the already covered positions.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"However, it is desirable if we can separate them.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If \u00e2\u0080\u009cgun\u00e2\u0080\u009d and \u00e2\u0080\u009crevolver\u00e2\u0080\u009d refer to the same object, then it should also be acceptable to say that Fred was \u00e2\u0080\u009ckilled with a gun\u00e2\u0080\u009d and that the burglar \u00e2\u0080\u009cfireda revolver\u00e2\u0080\u009d.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":", for A. T.&T. nonalpha.. .","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The maximum entropy framework estimates probabilities based on the principle of making as few assumptions as possible, other than the constraints imposed.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For example, one of the ATB samples was the determiner -\"\" ; dhalik\u00e2\u0080\u009cthat.\u00e2\u0080\u009d The sample occurred in 1507 corpus po sitions, and we found that the annotations were consistent.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"One hybridization strategy is to let the parsers vote on constituents' membership in the hypothesized set.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"From this we see that a finer-grained model for parser combination, at least for the features we have examined, will not give us any additional power.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"An analysis of nouns that occur in both the singular and the plural in our database reveals that there is indeed a slight but significant positive correlation-R2 = 0.20, p < 0.005; see Figure 6.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"This assumption, however, is not inherent to type-based tagging models.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(See also Wu and Fung [1994].)","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The theory has also been validated empirically.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"All four of the techniques studied result in parsing systems that perform better than any previously reported.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The problem can be represented as a graph with 2N vertices corresponding to the members of X1 and X2.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"9 65.5 46.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We model each parse as the decisions made to create it, and model those decisions as independent events.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For irt the Good-Turing estimate just discussed gives us an estimate of p(unseen(f,) I f,)-the probability of observing a previously unseen instance of a construction in ft given that we know that we have a construction in f,.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(b) F.i'JJI!","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Using the terminology of Kahane et al. (1998), we say that jedna is the syntactic head of Z, while je is its linear head in the projectivized representation.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The paper is structured as follows.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Also, the argument has been made that machine translation performance should be evaluated via task-based evaluation metrics, i.e. how much it assists performing a useful task, such as supporting human translators or aiding the analysis of texts.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"As we have said, parse quality decreases with sentence length.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"As the name implies, space is O(m) and linear in the number of entries.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"A corpus of German newspaper commentaries has been assembled at Potsdam University, and annotated with different linguistic information, to different degrees.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This would result in better rest cost estimation and better pruning.10 In general, tighter, but well factored, integration between the decoder and language model should produce a significant speed improvement.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Some of these approaches (e.g., Lin, Chiang, and Su [1993]) attempt to identify unknown words, but do not ac\u00c2\u00ad tually tag the words as belonging to one or another class of expression.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The compressed variant uses block compression and is rather slow as a result.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Mixing, smoothing, and instance-feature weights are learned at the same time using an efficient maximum-likelihood procedure that relies on only a small in-domain development corpus.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"3.1 General Knowledge Sources.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The returned state s(wn1) may then be used in a followon query p(wn+1js(wn1)) that extends the previous query by one word.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(In this figure eps is c) be implemented, though, such as a maximal-grouping strategy (as suggested by one reviewer of this paper); or a pairwise-grouping strategy, whereby long sequences of unattached hanzi are grouped into two-hanzi words (which may have some prosodic motivation).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"(2006).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"A better approach would be to distin guish between these cases, possibly by drawing on the vast linguistic work on Arabic connectives (AlBatal, 1990).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Thus we have some confidence that our own performance is at least as good as that of Chang et al.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We indicate whether a context with zero log backoff will extend using the sign bit: +0.0 for contexts that extend and \u22120.0 for contexts that do not extend.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"None of the models attach the attributive adjectives correctly.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Compared with the widely- SRILM, our is 2.4 times as fast while using 57% of the mem- The structure is a trie with bit-level packing, sorted records, interpolation search, and optional quantization aimed lower memory consumption. simultaneously uses less memory than the smallest lossless baseline and less CPU than the baseline.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"While the first three models get three to four tags wrong, our best model gets only one word wrong and is the most accurate among the four models for this example.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Sentence pairs are the natural instances for SMT, but sentences often contain a mix of domain-specific and general language.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Specifically, for the ith word type, the set of token-level tags associated with token occurrences of this word, denoted t(i), must all take the value Ti to have nonzero mass. Thus in the context of Gibbs sampling, if we want to block sample Ti with t(i), we only need sample values for Ti and consider this setting of t(i).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Section 2.1 describes how BABAR generates training examples to use in the learning process.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The attachment in such cases encompasses a long distance dependency that cannot be captured by Markovian processes that are typically used for morphological disambiguation.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2009).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"These estimates are in turn combined linearly with relative-frequency estimates from an in-domain phrase table.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"If the same pair of NE instances is used with different phrases, these phrases are likely to be paraphrases.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The structure uses linear probing hash tables and is designed for speed.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Two sets of examples from Gan are given in (1) and (2) (:::::: Gan's Appendix B, exx.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We have used the Java-based opennlp maximum entropy package1.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Timing is based on plentiful memory.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Given a document to process, BABAR uses four modules to perform coreference resolution.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"However, their system is a hybrid of hand-coded rules and machine learning methods.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"For each source word f, the list of its possible translations e is sorted according to p(fje) puni(e), where puni(e) is the unigram probability of the English word e. It is su\u00c3\u0086cient to consider only the best 50 words.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"An ATM may be thought of as spawning independent processes for each applicable move.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Except our own and MENE + reference resolution, the results in Table 6 are all official MUC7 results.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If it starts with a lower case letter, and contains both upper and lower case letters, then (mixedCaps, zone) is set to 1.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"However, to classify a token , while Borthwick uses tokens from to (from two tokens before to two tokens after ), we used only the tokens , , and . Even with local features alone, MENERGI outperforms MENE (Borthwick, 1999).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Obviously \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is part of the following clause rather than being the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d and the extracted instance makes no sense.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"If the log backoff of wnf is also zero (it may not be in filtered models), then wf should be omitted from the state.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"They are also labelled for their topicality (yes \/ no), and this annotation is accompanied by a confidence value assigned by the annotator (since it is a more subjective matter).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Jud ges A G G R ST M 1 M 2 M 3 T1 T2 T3 AG 0.7 0 0.7 0 0 . 4 3 0.4 2 0.6 0 0.6 0 0.6 2 0.5 9 GR 0.9 9 0 . 6 2 0.6 4 0.7 9 0.8 2 0.8 1 0.7 2 ST 0 . 6 4 0.6 7 0.8 0 0.8 4 0.8 2 0.7 4 M1 0.7 7 0.6 9 0.7 1 0.6 9 0.7 0 M2 0.7 2 0.7 3 0.7 1 0.7 0 M3 0.8 9 0.8 7 0.8 0 T1 0.8 8 0.8 2 T2 0.7 8 respectively, the recall and precision.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Gather phrases using keywords Next, we select a keyword for each phrase \u00e2\u0080\u0093 the top-ranked word based on the TF\/IDF metric.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For natural disasters, BABAR generated 20,479 resolutions: 11,652 from lexical seeding and 8,827 from syntactic seeding.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Our final average POS tagging accuracy of 83.4% compares very favorably to the average accuracy of Berg-Kirkpatrick et al.\u2019s monolingual unsupervised state-of-the-art model (73.0%), and considerably bridges the gap to fully supervised POS tagging performance (96.6%).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Consequently, all three parsers prefer the nominal reading.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Second, we treat the projected labels as features in an unsupervised model (\u00a75), rather than using them directly for supervised training.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In German, the verbgroup usually consists of a left and a right verbal brace, whereas in English the words of the verbgroup usually form a sequence of consecutive words.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The 95% confidence intervals for type-level errors are (5580, 9440) for the ATB and (1400, 4610) for the WSJ.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Instead, the designs of the various annotation layers and the actual annotation work are results of a series of diploma theses, of students\u00e2\u0080\u0099 work in course projects, and to some extent of paid assistentships.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"mark- ContainsVerb is especially effective for distinguishing root S nodes of equational sentences.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For RandLM, we used the settings in the documentation: 8 bits per value and false positive probability 1 256.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"All the NE pair instances which co-occur separated by at most 4 chunks are collected along with information about their NE types and the phrase between the NEs (the \u00e2\u0080\u0098context\u00e2\u0080\u0099).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"(Webber et al., 2003)).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We further thank Dr. J.-S.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Compared with the widely- SRILM, our is 2.4 times as fast while using 57% of the mem- The structure is a trie with bit-level packing, sorted records, interpolation search, and optional quantization aimed lower memory consumption. simultaneously uses less memory than the smallest lossless baseline and less CPU than the baseline.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Our empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"These methods demonstrated the benefits of incorporating linguistic features using a log-linear parameterization, but requires elaborate machinery for training.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We group the features used into feature groups.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"While the semantic aspect of radicals is by no means completely predictive, the semantic homogeneity of many classes is quite striking: for example 254 out of the 263 examples (97%) of the INSECT class listed by Wieger (1965, 77376) denote crawling or invertebrate animals; similarly 21 out of the 22 examples (95%) of the GHOST class (page 808) denote ghosts or spirits.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The computing time is low, since no reordering is carried out.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"As with HG's derivation structures are annotated; in the case of TAG's, by the trees used for adjunction and addresses of nodes of the elementary tree where adjunctions occurred.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"For verbs we add two features.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In our model, however, all lattice paths are taken to be a-priori equally likely.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The results in Table 2 were achieved on the development set.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"State is implemented in their scrolling variant, which is a trie annotated with forward and backward pointers.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The effect of UNIQ is very small on both data sets.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Given a key k, it estimates the position If the estimate is exact (A[pivot] = k), then the algorithm terminates succesfully.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The paper is organized as follows: Section 2 explains the different layers of annotation that have been produced or are being produced.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In the labeled version of these metrics (L) both heads and arc labels must be correct, while the unlabeled version (U) only considers heads.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Previous work that involves the gathering of information from the whole document often uses a secondary classifier, which corrects the mistakes of a primary sentence- based classifier.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Other work includes transferring latent topic distributions from source to target language for LM adaptation, (Tam et al., 2007) and adapting features at the sentence level to different categories of sentence (Finch and Sumita, 2008).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Ex: Mr. Cristiani is the president ...","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In (Bikel et al., 1997) and (Bikel et al., 1999), performance was plotted against training data size to show how performance improves with training data size.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In many cases these failures in recall would be fixed by having better estimates of the actual prob\u00c2\u00ad abilities of single-hanzi words, since our estimates are often inflated.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We present two algorithms.","label":1},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Similarly, hanzi sharing the GHOST radical _m tend to denote spirits and demons, such as _m gui3 'ghost' itself, II: mo2 'demon,' and yan3 'nightmare.'","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"It also does not prune, so comparing to our pruned model would be unfair.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"It's not clear how to apply these methods in the unsupervised case, as they required cross-validation techniques: for this reason we use the simpler smoothing method shown here. input to the unsupervised algorithm is an initial, &quot;seed&quot; set of rules.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The natural language processing community is in the strong position of having many available approaches to solving some of its most fundamental problems.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"raphy: A ren2 'person' is a fairly uncontroversial case of a monographemic word, and rplil zhong1guo2 (middle country) 'China' a fairly uncontroversial case of a di\u00c2\u00ad graphernic word.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We can 5 Recall that precision is defined to be the number of correct hits divided by the total number of items.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Another way to interpret this is that less than 5% of the correct constituents are missing from the hypotheses generated by the union of the three parsers.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In the cotraining case, (Blum and Mitchell 98) argue that the task should be to induce functions Ii and f2 such that So Ii and 12 must (1) correctly classify the labeled examples, and (2) must agree with each other on the unlabeled examples.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The difference in precision between similarity and Bayes switching techniques is significant, but the difference in recall is not.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Wu and Fung introduce an evaluation method they call nk-blind.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The PROBING data structure is a rather straightforward application of these hash tables to store Ngram language models.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Then each arc of D maps either from an element of H to an element of p, or from E-i.e., the empty string-to an element of P. More specifically, each word is represented in the dictionary as a sequence of arcs, starting from the initial state of D and labeled with an element 5 of Hxp, which is terminated with a weighted arc labeled with an element of Ex P. The weight represents the estimated cost (negative log probability) of the word.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Table 2 shows these similarity measures.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"When finished, the whole material is written into an XML-structured annotation file.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"As far as we know, no other NERs have used information from the whole document (global) as well as information within the same sentence (local) in one framework.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"With a good hash function, collisions of the full 64bit hash are exceedingly rare: one in 266 billion queries for our baseline model will falsely find a key not present.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(3) shows learning curves for CoBoost.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The major problem for all segmentation systems remains the coverage afforded by the dictionary and the lexical rules used to augment the dictionary to deal with unseen words.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Call the crossing constituents A and B.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Each name class is subdivided into 4 sub-classes, i.e., N begin, N continue, N end, and N unique.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The feature-based model replaces the emission distribution with a log-linear model, such that: on the word identity x, features checking whether x contains digits or hyphens, whether the first letter of x is upper case, and suffix features up to length 3.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"This work has been supported as part of the Verbmobil project (contract number 01 IV 601 A) by the German Federal Ministry of Education, Science, Research and Technology and as part of the Eutrans project (ESPRIT project number 30268) by the European Community.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"!!:\\ :yu2 e:_nc [::!!:zen3 l!f :moO t:_adv il!:shuot ,:_vb i i i 1 \u00e2\u0080\u00a2 10.03 13...","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Their work used subject-verb, verb-object, and adjective-noun relations to compare the contexts surrounding an anaphor and candidate.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"However, it is also mostly political content (even if not focused on the internal workings of the European Union) and opinion.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We used this data to build an unpruned ARPA file with IRSTLM\u2019s improved-kneser-ney option and the default three pieces.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For speed, we plan to implement the direct-mapped cache from BerkeleyLM.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The traveling salesman problem is an optimization problem which is defined as follows: given are a set of cities S = s1; ; sn and for each pair of cities si; sj the cost dij > 0 for traveling from city si to city sj . We are looking for the shortest tour visiting all cities exactly once while starting and ending in city s1.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Lexicon Feature: The string of the token is used as a feature.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Several papers report the use of part-of-speech information to rank segmentations (Lin, Chiang, and Su 1993; Peng and Chang 1993; Chang and Chen 1993); typically, the probability of a segmentation is multiplied by the probability of the tagging(s) for that segmentation to yield an estimate of the total probability for the analysis.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(2) was extended to have an additional, innermost loop over the (3) possible labels.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(7) is at 0 when: 1) Vi : sign(gi (xi)) = sign(g2 (xi)); 2) Ig3(xi)l oo; and 3) sign(gi (xi)) = yi for i = 1, , m. In fact, Zco provides a bound on the sum of the classification error of the labeled examples and the number of disagreements between the two classifiers on the unlabeled examples.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"As discussed in more detail in \u00a73, we use two types of vertices in our graph: on the foreign language side vertices correspond to trigram types, while the vertices on the English side are individual word types.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In the next section, we show how an ATM can accept the strings generated by a grammar in a LCFRS formalism in logspace, and hence show that each family can be recognized in polynomial time.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The learning task is to find two classifiers : 2x1 { \u20141, +1} 12 : 2x2 { \u20141, +1} such that (x1,) = f2(x2,t) = yt for examples i = 1, , m, and f1 (x1,) = f2 (x2,t) as often as possible on examples i = m + 1, ,n. To achieve this goal we extend the auxiliary function that bounds the training error (see Equ.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"18 We are grateful to ChaoHuang Chang for providing us with this set.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Lossy compressed models RandLM (Talbot and Osborne, 2007) and Sheffield (Guthrie and Hepple, 2010) offer better memory consumption at the expense of CPU and accuracy.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"(Kehler, 1997) also used a DempsterShafer model to merge evidence from different sources for template-level coreference.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The type-level posterior term can be computed according to, P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2) \u00e2\u0088\u009d Note that each round of sampling Ti variables takes time proportional to the size of the corpus, as with the standard token-level HMM.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The approach recursively evaluates a quantity Q(C; j), where C is the set of already visited cities and sj is the last visited city.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"needs to be in initCaps to be considered for this feature.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Language models are widely applied in natural language processing, and applications such as machine translation make very frequent queries.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Since the parameter and token components will remain fixed throughout experiments, we briefly describe each.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Recall.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"A Hebrew surface token may have several readings, each of which corresponding to a sequence of segments and their corresponding PoS tags.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"\"c' 0 + 0 \"0 ' \u00e2\u0080\u00a2 + a n t i g r e e d y x g r e e d y < > c u r r e n t m e t h o d o d i e t . o n l y \u00e2\u0080\u00a2 Taiwan 0 \u00c2\u00b7;; 0 c CD E i5 0\"' 9 9 \u00e2\u0080\u00a2 Mainland \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 -0.30.20.1 0.0 0.1 0.2 Dimension 1 (62%) Figure 7 Classical metric multidimensional scaling of distance matrix, showing the two most significant dimensions.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"If either case is true, then CFLex reports that the anaphor and candidate might be coreferent.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We can now add a new weak hypothesis 14 based on a feature in X1 with a confidence value al hl and atl are chosen to minimize the function We now define, for 1 <i <n, the following virtual distribution, As before, Ztl is a normalization constant.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"So we decided to use semantic class information only to rule out candidates.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Across all languages, high performance can be attained by selecting a single tag per word type.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"BerkeleyLM revision 152 (Pauls and Klein, 2011) implements tries based on hash tables and sorted arrays in Java with lossy quantization.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The quasi-monotone search performs best in terms of both error rates mWER and SSER.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"37 84.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The problem of &quot;noise&quot; items that do not fall into any of the three categories also needs to be addressed.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We used it to score all phrase pairs in the OUT table, in order to provide a feature for the instance-weighting model.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Performance improvements transfer to the Moses (Koehn et al., 2007), cdec (Dyer et al., 2010), and Joshua (Li et al., 2009) translation systems where our code has been integrated.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Most similar is scrolling queries, wherein left-to-right queries that add one word at a time are optimized.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Both implementations employ a state object, opaque to the application, that carries information from one query to the next; we discuss both further in Section 4.2.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We also report word type level accuracy, the fraction of word types assigned their majority tag (where the mapping between model state and tag is determined by greedy one-to-one mapping discussed above).5 For each language, we aggregate results in the following way: First, for each hyperparameter setting, evaluate three variants: The first model (1TW) only 4 Typically, the performance stabilizes after only 10 itera-.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"All notions of word, with the exception of the orthographic word, are as relevant in Chinese as they are in English, and just as is the case in other languages, a word in Chinese may correspond to one or more symbols in the orthog 1 For a related approach to the problem of word-segrnention in Japanese, see Nagata (1994), inter alia..","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Input: Ja , wunderbar . Konnen wir machen . MonS: Yes, wonderful.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"This feature includes named entities, which the ATB marks with a flat NP node dominating an arbitrary number of NNP pre-terminal daughters (Figure 2).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Although these existential NPs do not need a prior referent, they may occur multiple times in a document.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We define the lattice L to be the concatenation of the lattices Li corresponding to the input words wi (s.t.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"A novel element of our model is the ability to capture type-level tag frequencies.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"By making use of global context, it has achieved excellent results on both MUC6 and MUC7 official test data.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Le m\u00b4edicament de r\u00b4ef\u00b4erence de Silapo est EPREX\/ERYPO, qui contient de l\u2019\u00b4epo\u00b4etine alfa.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"In a model we built with default settings, 1.2% of n + 1-grams were missing their ngram suffix.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"They cluster NE instance pairs based on the words in the contexts using a bag- of-words method.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The substructures that are unanimously hypothesized by the parsers should be preserved after combination, and the combination technique should not foolishly create substructures for which there is no supporting evidence.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"IRSTLM (Federico et al., 2008) is an open-source toolkit for building and querying language models.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"ATB CTB6 Negra WSJ Trees 23449 28278 20602 43948 Word Typess 40972 45245 51272 46348 Tokens 738654 782541 355096 1046829 Tags 32 34 499 45 Phrasal Cats 22 26 325 27 Test OOV 16.8% 22.2% 30.5% 13.2% Per Sentence Table 4: Gross statistics for several different treebanks.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"For parsing, the most challenging form of ambiguity occurs at the discourse level.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We received submissions from 14 groups from 11 institutions, as listed in Figure 2.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The adjunction operation with respect to tree sets (multicomponent adjunction) is defined as follows.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Recently, statistical NERs have achieved results that are comparable to hand-coded systems.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Our approach uses word-to-word dependencies between source and target words.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(3) shows learning curves for CoBoost.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"In this particular case, all English vertices are labeled as nouns by the supervised tagger.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Our representation of contextual roles is based on information extraction patterns that are converted into simple caseframes.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"3.2 Stochastic rhetorical analysis.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"For example, in the phrase \u00e2\u0080\u009cCompany-A last week purchased rival Marshalls from Company-B\u00e2\u0080\u009d, the purchased company is Marshalls, not Company-B.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"of Arabic.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This group consists of (3 total number of possible zones) features.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"72 77.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The first row represents the average accuracy of the three parsers we combine.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Proper names that match are resolved with each other.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We again adopt an approach where we alternate between two classifiers: one classifier is modified while the other remains fixed.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"newspaper material, but also including kungfu fiction, Buddhist tracts, and scientific material.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Many morphological decisions are based on long distance dependencies, and when the global syntactic evidence disagrees with evidence based on local linear context, the two models compete with one another, despite the fact that the PCFG takes also local context into account.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Yet we note that the better grammars without pruning outperform the poorer grammars using this technique, indicating that the syntactic context aids, to some extent, the disambiguation of unknown tokens.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The natural baseline approach is to concatenate data from IN and OUT.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"JA DE G O L D G R AS S SI C K NE SS DE AT H R A T 14.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Nonetheless, the prospect of a network of annotated discourse resources seems particularly promising if not only a single annotation layer is used but a whole variety of them, so that a systematic search for correlations between them becomes possible, which in turn can lead to more explanatory models of discourse structure.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We used a standard one-pass phrase-based system (Koehn et al., 2003), with the following features: relative-frequency TM probabilities in both directions; a 4-gram LM with Kneser-Ney smoothing; word-displacement distortion model; and word count.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Given that part-of-speech labels are properties of words rather than morphemes, it follows that one cannot do part-of-speech assignment without having access to word-boundary information.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We characterize a class of formalisms that have this property in Section 4.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"But gold segmentation is not available in application settings, so a segmenter and parser are arranged in a pipeline.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"For illustration, an English translation of one of the commentaries is given in Figure 1.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Figure 2 shows a screenshot (which is of somewhat limited value, though, as color plays a major role in signalling the different statuses of the information).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The token-level term is similar to the standard HMM sampling equations found in Johnson (2007).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For example, as Gan (1994) has noted, one can construct examples where the segmen\u00c2\u00ad tation is locally ambiguous but can be determined on the basis of sentential or even discourse context.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Subsets C of increasing cardinality c are processed.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Arguably this consists of about three phonological words.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Information from a sentence is sometimes insufficient to classify a name correctly.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"During development, we sensed that the Recency and Syn- role KSs did not deserve to be on equal footing with the other KSs because their knowledge was so general.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"TIS systems in general need to do more than simply compute the.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The computing time is given in terms of CPU time per sentence (on a 450MHz PentiumIIIPC).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The confidence intervals are computed by bootstrap resampling for BLEU, and by standard significance testing for the manual scores, as described earlier in the paper.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We found that contextual role knowledge was more beneficial for pronouns than for definite noun phrases.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We excluded these from the evaluation as they can be easily identified with a list of days\/months.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In this section, we brie y review our translation approach.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This is an issue that we have not addressed at the current stage of our research.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Reiche\u00e2\u0080\u0099s colleagues will make sure that the concept is waterproof.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The same form fmnh can be segmented as f-mnh, f (\u201cthat\u201d) functioning as a reletivizer with the form mnh.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(a) Of the high frequency phrasal categories, ADJP and SBAR are the hardest to parse.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Not surprisingly some semantic classes are better for names than others: in our corpora, many names are picked from the GRASS class but very few from the SICKNESS class.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Feature weights were set using Och\u2019s MERT algorithm (Och, 2003).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, the characterization given in the main body of the text is correct sufficiently often to be useful.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"A Stochastic Finite-State Word-Segmentation Algorithm for Chinese","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Beyond optimizing the memory size of TRIE, there are alternative data structures such as those in Guthrie and Hepple (2010).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Our work is closest to that of Yarowsky and Ngai (2001), but differs in two important ways.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Therefore we opted to use the much 7 The DempsterShafer theory assumes that one of the hypotheses in \u00ce\u00b8 is correct, so eliminating all of the hypotheses violates this assumption.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In MUC6 and MUC7, the named entity task is defined as finding the following classes of names: person, organization, location, date, time, money, and percent (Chinchor, 1998; Sundheim, 1995) Machine learning systems in MUC6 and MUC 7 achieved accuracy comparable to rule-based systems on the named entity task.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The 14 general-language features embody straightforward cues: frequency, \u201ccentrality\u201d as reflected in model scores, and lack of burstiness.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For example, we can easily imagine that the number of paraphrases for \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d is enormous and it is not possible to create a comprehensive inventory by hand.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Indeed there are several open issues.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"35 76.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Limitations of (Blum and Mitchell 98): While the assumptions of (Blum and Mitchell 98) are useful in developing both theoretical results and an intuition for the problem, the assumptions are quite limited.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Table 1 The cost as a novel given name (second position) for hanzi from various radical classes.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"0 Figure 5 An example of affixation: the plural affix.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Presence of the determiner J Al. 2.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Similar behavior is observed when adding features.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Fortunately, there are only a few hundred hanzi that are particularly common in transliterations; indeed, the commonest ones, such as E. bal, m er3, and iij al are often clear indicators that a sequence of hanzi containing them is foreign: even a name like !:i*m xia4mi3-er3 'Shamir,' which is a legal Chi\u00c2\u00ad nese personal name, retains a foreign flavor because of liM.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Despite their simplicity, uni- gram weights have been shown as an effective feature in segmentation models (Dyer, 2009).13 The joint parser\/segmenter is compared to a pipeline that uses MADA (v3.0), a state-of-the-art Arabic segmenter, configured to replicate ATB segmentation (Habash and Rambow, 2005).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Finally, we find links between sets of phrases, based on the NE instance pair data (for example, different phrases which link \u00e2\u0080\u009cIBM\u00e2\u0080\u009d and \u00e2\u0080\u009cLotus\u00e2\u0080\u009d) (Step 4).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We will briefly discuss this point in Section 3.1.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"There are two key benefits of this model architecture.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The difference between the featureless model (+PRIOR) and our full model (+FEATS) is 13.6% and 7.7% average error reduction on best and median settings respectively.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Kollege.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In addition to a heuristic based on decision list learning, we also presented a boosting-like framework that builds on ideas from (Blum and Mitchell 98).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"(S; C; j); Not only the coverage set C and the positions j; j0, but also the verbgroup states S; S0 are taken into account.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"(2010) and the posterior regular- ization HMM of Grac\u00c2\u00b8a et al.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"One is the accuracy within a set of phrases which share the same keyword; the other is the accuracy of links.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As the reviewer also points out, this is a problem that is shared by, e.g., probabilistic context-free parsers, which tend to pick trees with fewer nodes.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Note that it is in precision that our over\u00c2\u00ad all performance would appear to be poorer than the reported performance of Chang et al., yet based on their published examples, our system appears to be doing better precisionwise.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"While there might be some controversy about the exact definition of such a tagset, these 12 categories cover the most frequent part-of-speech and exist in one form or another in all of the languages that we studied.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(1991}, Gu and Mao (1994), and Nie, Jin, and Hannan (1994).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"10) and trained both EM and L-BFGS for 1000 iterations.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In many cases, there is an even stronger restriction: over large portions of the source string, the alignment is monotone.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"It has been shown for English (Wang and Hirschberg 1992; Hirschberg 1993; Sproat 1994, inter alia) that grammatical part of speech provides useful information for these tasks.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Fort= 1,...,T:","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Let us notate the set of previously unseen, or novel, members of a category X as unseen(X); thus, novel members of the set of words derived in f, menO will be de\u00c2\u00ad noted unseen(f,).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"However, their system is a hybrid of hand-coded rules and machine learning methods.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"For all languages we do not make use of a tagging dictionary.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In a more recent study than Chang et al., Wang, Li, and Chang (1992) propose a surname-driven, non-stochastic, rule-based system for identifying personal names.17 Wang, Li, and Chang also compare their performance with Chang et al.'s system.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"As we will see from Table 3, not much improvement is derived from this feature.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Up to now, most IE researchers have been creating paraphrase knowledge (or IE patterns) by hand and for specific tasks.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Figure 2: An ATB sample from the human evaluation.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Figure 3 shows examples of semantic expectations that were learned.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The effect of the pruning threshold t0 is shown in Table 5.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Our model outperforms theirs on four out of five languages on the best hyperparameter setting and three out of five on the median setting, yielding an average absolute difference across languages of 12.9% and 3.9% for best and median settings respectively compared to their best EM or LBFGS performance.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"For our demonstration system, we typically use the pruning threshold t0 = 5:0 to speed up the search by a factor 5 while allowing for a small degradation in translation accuracy.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Quantization can be improved by jointly encoding probability and backoff.","label":0},{"source_text":"There is no global pruning.","target_text":"f1; ; Jg denotes a coverage set including all positions from the starting position 1 to position J and j 2 fJ \u00f4\u0080\u0080\u0080L; ; Jg.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"There are 13 types of extensions needed to describe the verbgroup reordering.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Time starts when Moses is launched and therefore includes model loading time.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"If the context wnf will never extend to the right (i.e. wnf v is not present in the model for all words v) then no subsequent query will match the full context.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Step 3.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In this section, we try to compare our results with those obtained by IdentiFinder ' 97 (Bikel et al., 1997), IdentiFinder ' 99 (Bikel et al., 1999), and MENE (Borthwick, 1999).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The accuracy is calculated as the ratio of the number of paraphrases to the total number of phrases in the set.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Although the segmentation requirements for Arabic are not as extreme as those for Chinese, Arabic is written with certain cliticized prepositions, pronouns, and connectives connected to adjacent words.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"32 81.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Using less training data than other systems, our NER is able to perform as well as other state-of-the-art NERs.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"RandLM and SRILM also remove context that will not extend, but SRILM performs a second lookup in its trie whereas our approach has minimal additional cost.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The form fmnh, for example, can be understood as the verb \u201clubricated\u201d, the possessed noun \u201cher oil\u201d, the adjective \u201cfat\u201d or the verb \u201cgot fat\u201d.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The sentence-selection approach is crude in that it imposes a binary distinction between useful and non-useful parts of OUT.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Although the tag distributions of the foreign words (Eq.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This heuristics is used to prune all segmentation possibilities involving \u201clexically improper\u201d segments.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Modifying the Berkeley parser for Arabic is straightforward.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We thank members of the MIT NLP group for their suggestions and comments.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In considering the recognition of these languages, we were forced to be more specific regarding the relationship between the structures derived by these formalisms and the substrings they span.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Each feature concept is akin to a random variable and its occurrence in the text corresponds to a particular instantiation of that random variable.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If the semantic expectations do not intersect, then we know that the case- frames extract mutually exclusive types of noun phrases.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Top 5 Bot to m 5 Go ld NN P NN JJ CD NN S RB S PD T # \u00e2\u0080\u009d , 1T W CD W RB NN S VB N NN PR P$ W DT : MD . +P RI OR CD JJ NN S WP $ NN RR B- , $ \u00e2\u0080\u009d . +F EA TS JJ NN S CD NN P UH , PR P$ # . \u00e2\u0080\u009c Table 5: Type-level English POS Tag Ranking: We list the top 5 and bottom 5 POS tags in the lexicon and the predictions of our models under the best hyperparameter setting.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"21 In Chinese, numerals and demonstratives cannot modify nouns directly, and must be accompanied by.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Even without features, but still using the tag prior, our median result is 52.0%, still significantly outperforming Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"In order to evaluate and advance this approach, it helps to feed into the knowledge base data that is already enriched with some of the desired information \u00e2\u0080\u0094 as in PCC.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Word type N % Dic tion ary entr ies 2 , 5 4 3 9 7 . 4 7 Mor pho logi call y deri ved wor ds 3 0 . 1 1 Fore ign tran slite rati ons 9 0 . 3 4 Per son al na mes 5 4 2 . 0 7 cases.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Furthermore, some of the arcs represent lexemes not present in the input tokens (e.g. h\/DT, fl\/POS), however these are parts of valid analyses of the token (cf. super-segmental morphology section 2).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"We will show that by giving the first model some global features, MENERGI outperforms Borthwick' s reference resolution classifier.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"They contain about 200M words (25M, 110M, 40M and 19M words, respectively).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Arguably this consists of about three phonological words.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We allow any number of bits from 2 to 25, unlike IRSTLM (8 bits) and BerkeleyLM (17\u221220 bits).","label":0},{"source_text":"There is no global pruning.","target_text":"Here, $ is the sentence boundary symbol, which is thought to be at position 0 in the target sentence.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"92 77.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Fourth, we show how to build better models for three different parsers.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Figure 2 shows an excerpt of a sentence from the Italian test set and the tags assigned by four different models, as well as the gold tags.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The work of the first author was supported by the Lynn and William Frankel Center for Computer Sciences.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"If we were working with more than three parsers we could investigate minority constituents, those constituents that are suggested by at least one parser, but which the majority of the parsers do not suggest.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Certainly these linguistic factors increase the difficulty of syntactic disambiguation.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"7 Conclusion and Future Work.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Oracle results).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"In our model, we associate these features at the type-level in the lexicon.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The NIST06 and NIST08 evaluation sets were used for testing.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The natural language processing community is in the strong position of having many available approaches to solving some of its most fundamental problems.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"In the rhetorical tree, nuclearity information is then used to extract a \u00e2\u0080\u009ckernel tree\u00e2\u0080\u009d that supposedly represents the key information from which the summary can be generated (which in turn may involve co-reference information, as we want to avoid dangling pronouns in a summary).","label":0},{"source_text":"There is no global pruning.","target_text":"Our approach uses word-to-word dependencies between source and target words.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In the numerator, however, the counts of ni1s are quite irregular, in\u00c2\u00ad cluding several zeros (e.g., RAT, none of whose members were seen).","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The choice between different actions is in general nondeterministic, and the parser relies on a memorybased classifier, trained on treebank data, to predict the next action based on features of the current parser configuration.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Preferences for constituent order (especially in languages with relatively free word order) often belong to this group.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The problem is to store these two values for a large and sparse set of n-grams in a way that makes queries efficient.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The work of the first author was supported by the Lynn and William Frankel Center for Computer Sciences.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Mixing, smoothing, and instance-feature weights are learned at the same time using an efficient maximum-likelihood procedure that relies on only a small in-domain development corpus.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"For humans, this characteristic can impede the acquisition of literacy.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In principle a feature could be an arbitrary predicate of the (spelling, context) pair; for reasons that will become clear, features are limited to querying either the spelling or context alone.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If somewhere else in the document we see \u00e2\u0080\u009crestrictions put in place by President Bush\u00e2\u0080\u009d, then we can be surer that Bush is a name.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"To conserve memory at the expense of accuracy, values may be quantized using q bits per probability and r bits per backoff6.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The method shares some characteristics of the decision list algorithm presented in this paper.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The data is sorted based on the frequency of the context (\u00e2\u0080\u009ca unit of\u00e2\u0080\u009d appeared 314 times in the corpus) and the NE pair instances appearing with that context are shown with their frequency (e.g. \u00e2\u0080\u009cNBC\u00e2\u0080\u009d and \u00e2\u0080\u009cGeneral Electric Co.\u00e2\u0080\u009d appeared 10 times with the context \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"This result suggests the benefit of using the automatic discovery method.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"First, in section 4, we evaluate the graph transformation techniques in themselves, with data from the Prague Dependency Treebank and the Danish Dependency Treebank.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"An analysis of nouns that occur in both the singular and the plural in our database reveals that there is indeed a slight but significant positive correlation-R2 = 0.20, p < 0.005; see Figure 6.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"There has been additional recent work on inducing lexicons or other knowledge sources from large corpora.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The state function is integrated into the query process so that, in lieu of the query p(wnjwn\u22121 1 ), the application issues query p(wnjs(wn\u22121 1 )) which also returns s(wn1 ).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Note that the good classes JADE, GOLD and GRASS have lower costs than the bad classes SICKNESS, DEATH and RAT, as desired, so the trend observed for the results of this method is in the right direction.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The tool we use is MMAX8, which has been specifically designed for marking co-reference.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The entries in such a lexicon may be thought of as meaningful surface segments paired up with their PoS tags li = (si, pi), but note that a surface segment s need not be a space-delimited token.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We thank members of the MIT NLP group for their suggestions and comments.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"To measure the contribution of each modification, a third, intermediate algorithm, Yarowsky-cautious was also tested.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Semantic expectations are analogous to lexical expectations except that they represent semantic classes rather than nouns.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"As with lexical expections, the semantic classes of co-referring expressions are 4 They may not be perfectly substitutable, for example one NP may be more specific (e.g., \u00e2\u0080\u009che\u00e2\u0080\u009d vs. \u00e2\u0080\u009cJohn F. Kennedy\u00e2\u0080\u009d).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The baseline system in Table 3 refers to the maximum entropy system that uses only local features.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In (b) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the kidnapping victims, but in (c) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the armed men.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"It did 402 queries\/ms using 1.80 GB. cMemory use increased during scoring due to batch processing (MIT) or caching (Rand).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Context from the whole document can be important in classifying a named entity.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The choice of the genre commentary resulted from the fact that an investigation of rhetorical structure, its interaction with other aspects of discourse structure, and the prospects for its automatic derivation are the key motivations for building up the corpus.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"At this step, we will try to link those sets, and put them into a single cluster.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We adopted this state-of-the-art model because it makes it easy to experiment with various ways of incorporating our novel constraint feature into the log-linear emission model.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"First we mark any node that dominates (at any level) a verb sider POS tags when pre-terminals are the only intervening nodes between the nucleus and its bracketing (e.g., unaries, base NPs).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The features we used can be divided into 2 classes: local and global.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Fortunately, there are only a few hundred hanzi that are particularly common in transliterations; indeed, the commonest ones, such as E. bal, m er3, and iij al are often clear indicators that a sequence of hanzi containing them is foreign: even a name like !:i*m xia4mi3-er3 'Shamir,' which is a legal Chi\u00c2\u00ad nese personal name, retains a foreign flavor because of liM.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This is orthographically represented as 7C.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"This vector tx is constructed for every word in the foreign vocabulary and will be used to provide features for the unsupervised foreign language POS tagger.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"SEGTok(noH) is the segmentation accuracy ignoring mistakes involving the implicit definite article h.11 To evaluate our performance on the tagging task we report CPOS and FPOS corresponding to coarse- and fine-grained PoS tagging results (F1) measure.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The P (T |\u00cf\u0088) distribution, in English for instance, should have very low mass for the DT (determiner) tag, since determiners are a very small portion of the vocabulary.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":".","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Our use of an unweighted lattice reflects our belief that all the segmentations of the given input sentence are a-priori equally likely; the only reason to prefer one segmentation over the another is due to the overall syntactic context which is modeled via the PCFG derivations.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"3 61.7 38.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"A configuration of M consists of a state of the finite control, the nonblank contents of the input tape and k work tapes, and the position of each head.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In our experiments we set the parameter values randomly, and then ran EM to convergence.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For a trigram language model, the partial hypotheses are of the form (e0; e; C; j).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Hence, we take the probability of the event fmnh analyzed as REL VB to be This means that we generate f and mnh independently depending on their corresponding PoS tags, and the context (as well as the syntactic relation between the two) is modeled via the derivation resulting in a sequence REL VB spanning the form fmnh. based on linear context.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In (1) the sequencema3lu4 cannot be resolved locally, but depends instead upon broader context; similarly in (2), the sequence :::tcai2neng2 cannot be resolved locally: 1.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Hence we decided to restrict ourselves to only information from the same document.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"To set \u03b2, we used the same criterion as for \u03b1, over a dev corpus: The MAP combination was used for TM probabilities only, in part due to a technical difficulty in formulating coherent counts when using standard LM smoothing techniques (Kneser and Ney, 1995).3 Motivated by information retrieval, a number of approaches choose \u201crelevant\u201d sentence pairs from OUT by matching individual source sentences from IN (Hildebrand et al., 2005; L\u00a8u et al., 2007), or individual target hypotheses (Zhao et al., 2004).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The interdependence between fb or 1\/!i, and 5:2 is not captured by our model, but this could easily be remedied.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Furthermore, even the size of the dictionary per se is less important than the appropriateness of the lexicon to a particular test corpus: as Fung and Wu (1994) have shown, one can obtain substantially better segmentation by tailoring the lexicon to the corpus to be segmented.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Moreover, if we want to test on a huge test corpus, indexing the whole corpus might prove computationally expensive.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"3.2 Inter-annotator Agreement.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"0 70.9 42.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Both parameters depend on a single hyperparameter \u00ce\u00b1.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"To bridge this gap, we consider a practically motivated scenario, in which we want to leverage existing resources from a resource-rich language (like English) when building tools for resource-poor foreign languages.1 We assume that absolutely no labeled training data is available for the foreign language of interest, but that we have access to parallel data with a resource-rich language.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The range of the score is between 0 and 1 (higher is better).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Linear probing places at most one entry in each bucket.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The choice of the particular newspaper was motivated by the fact that the language used in a regional daily is somewhat simpler than that of papers read nationwide.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Each model was able to produce hypotheses for all input sentences.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"However, such an approach requires a number of hand-coded systems, which may not be available in languages other than English.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Any opinions, findings, conclusions, or recommendations expressed in this paper are those of the authors, and do not necessarily reflect the views of the funding organizations.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This analyzer setting is similar to that of (Cohen and Smith, 2007), and models using it are denoted nohsp, Parser and Grammar We used BitPar (Schmid, 2004), an efficient general purpose parser,10 together with various treebank grammars to parse the input sentences and propose compatible morphological segmentation and syntactic analysis.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The complexity of the algorithm is O(E3 J2 2J), where E is the size of the target language vocabulary.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Another technique for parse hybridization is to use a na\u00efve Bayes classifier to determine which constituents to include in the parse.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"3 61.7 38.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"(f1; ;mg n fl1; l2; l3g ;m) German to English the monotonicity constraint is violated mainly with respect to the German verbgroup.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"m(S) represents the belief that the correct hypothesis is included in S. The model assumes that evidence also arrives as a probability density function (pdf) over sets of hypotheses.6 Integrating new evidence into the existing model is therefore simply a matter of defining a function to merge pdfs, one representing the current belief system and one representing the beliefs of the new evidence.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Smith estimates Lotus will make profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For a sequence of hanzi that is a possible name, we wish to assign a probability to that sequence qua name.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The maximum precision row is the upper bound on accuracy if we could pick exactly the correct constituents from among the constituents suggested by the three parsers.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Finally, this effort is part of a much larger program that we are undertaking to develop stochastic finite-state methods for text analysis with applications to TIS and other areas; in the final section of this paper we will briefly discuss this larger program so as to situate the work discussed here in a broader context.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Since foreign names can be of any length, and since their original pronunciation is effectively unlimited, the identi\u00c2\u00ad fication of such names is tricky.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Consequently, all three parsers prefer the nominal reading.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The input to AdaBoost is a set of training examples ((xi , yi), , (x\u201e.\u201e yrn)).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Cohen and Smith (2007) followed up on these results and proposed a system for joint inference of morphological and syntactic structures using factored models each designed and trained on its own.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Our TRIE implements the popular reverse trie, in which the last word of an n-gram is looked up first, as do SRILM, IRSTLM\u2019s inverted variant, and BerkeleyLM except for the scrolling variant.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The two knowledge sources that use semantic expectations, WordSemCFSem and CFSemCFSem, always return values of -1 or 0.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"It is natural that the larger the data in the domain, the more keywords are found.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Schapire and Singer show that the training error is bounded above by Thus, in order to greedily minimize an upper bound on training error, on each iteration we should search for the weak hypothesis ht and the weight at that minimize Z.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"It is a sequence of proper nouns within an NP; its last word Cooper is the head of the NP; and the NP has an appositive modifier (a vice president at S.&P.) whose head is a singular noun (president).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The use of ILP in learning the desired grammar significantly increases the computational complexity of this method.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Memory mapping also allows the same model to be shared across processes on the same machine.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"76 16.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"To formalize the approach, we introduce four verbgroup states S: Initial (I): A contiguous, initial block of source positions is covered.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"That is, we can use the discourse parser on PCC texts, emulating for instance a \u00e2\u0080\u009cco-reference oracle\u00e2\u0080\u009d that adds the information from our co-reference annotations.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We generate these caseframes automatically by running AutoSlog over the training corpus exhaustively so that it literally generates a pattern to extract every noun phrase in the corpus.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"A configuration of M consists of a state of the finite control, the nonblank contents of the input tape and k work tapes, and the position of each head.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Unfortunately, there is no standard corpus of Chinese texts, tagged with either single or multiple human judgments, with which one can compare performance of various methods.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Evaluation Metrics We report three metrics to evaluate tagging performance.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"(Hartmann 1984), for example, used the term Reliefgebung to characterize the distibution of main and minor information in texts (similar to the notion of nuclearity in RST).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Evaluation of links A link between two sets is considered correct if the majority of phrases in both sets have the same meaning, i.e. if the link indicates paraphrase.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"AdaBoost was first introduced in (Freund and Schapire 97); (Schapire and Singer 98) gave a generalization of AdaBoost which we will use in this paper.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Consequently, we cut their evidence values in half to lessen their influence.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In addition, there are several approaches to non-projective dependency parsing that are still to be evaluated in the large (Covington, 1990; Kahane et al., 1998; Duchier and Debusmann, 2001; Holan et al., 2001; Hellwig, 2003).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"30 75.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"For all these annotation taks, G\u00c2\u00a8otze developed a series of questions (essentially a decision tree) designed to lead the annotator to the ap propriate judgement.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In previous work (Bean and Riloff, 1999), we developed an unsupervised learning algorithm that automatically recognizes definite NPs that are existential without syntactic modification because their meaning is universally understood.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The first point we need to address is what type of linguistic object a hanzi repre\u00c2\u00ad sents.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The TRIE model continues to use the least memory of ing (-P) with MAP POPULATE, the default.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Since the lattice L for a given sentence W is determined by the morphological analyzer M we have which is precisely the formula corresponding to the so-called lattice parsing familiar from speech recognition.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We do not adapt the alignment procedure for generating the phrase table from which the TM distributions are derived.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Using cn to denote the number of n-grams, total memory consumption of TRIE, in bits, is plus quantization tables, if used.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Formally, the DempsterShafer theory defines a probability density function m(S), where S is a set of hypotheses.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Two annotators received training with the RST definitions and started the process with a first set of 10 texts, the results of which were intensively discussed and revised.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We have used the Java-based opennlp maximum entropy package1.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"!!:\\ :yu2 e:_nc [::!!:zen3 l!f :moO t:_adv il!:shuot ,:_vb i i i 1 \u00e2\u0080\u00a2 10.03 13...","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This variant is tested in the experiments below.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Compared to the EMEA\/EP setting, the two domains in the NIST setting are less homogeneous and more similar to each other; there is also considerably more IN text available.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"To this end, we construct a bilingual graph over word types to establish a connection between the two languages (\u00a73), and then use graph label propagation to project syntactic information from English to the foreign language (\u00a74).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Figure 2 shows examples of extracted NE pair instances and their contexts.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This Good\u00c2\u00ad Turing estimate of p(unseen(f,) If,) can then be used in the normal way to define the probability of finding a novel instance of a construction in ir, in a text: p(unseen(f,)) = p(unseen(f,) I f,) p(fn Here p(ir,) is just the probability of any construction in ft as estimated from the frequency of such constructions in the corpus.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"First of all, most previous articles report perfor\u00c2\u00ad mance in terms of a single percent-correct score, or else in terms of the paired measures of precision and recall.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For 2 < n < N, we use a hash table mapping from the n-gram to the probability and backoff3.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"mark- ContainsVerb is especially effective for distinguishing root S nodes of equational sentences.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Ignoring the identity of the target language words e and e0, the possible partial hypothesis extensions due to the IBM restrictions are shown in Table 2.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective treebanks is often neglected because of the relative scarcity of problematic constructions.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Their work used subject-verb, verb-object, and adjective-noun relations to compare the contexts surrounding an anaphor and candidate.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"2.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Particles are uninflected.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We pick the parse that is most similar to the other parses by choosing the one with the highest sum of pairwise similarities.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"com t 600 Mountain Avenue, 2c278, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"In each figure the upper graph shows the isolated constituent precision and the bottom graph shows the corresponding number of hypothesized constituents.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Each parse is converted into a set of constituents represented as a tuples: (label, start, end).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We call such a constituent an isolated constituent.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"7).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"If these sets do not overlap, then the words cannot be coreferent.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The algorithm in Fig.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"more frequently than is done in English.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Gazdar (1985) considers a restriction of IG's in which no more than one nonterminal on the right-hand-side of a production can inherit the stack from the left-hand-side.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The numbers falling into the location, person, organization categories were 186, 289 and 402 respectively.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"However, lexically similar NPs usually refer to the same entity in two cases: proper names and existential noun phrases.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Acknowledgments We thank Steven Bethard, Evan Rosen, and Karen Shiells for material contributions to this work.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Particles are uninflected.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"(2010)\u00e2\u0080\u0099s richest model: optimized via either EM or LBFGS, as their relative performance depends on the language.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"As already pointed out in Section 2.4, current theories diverge not only on the number and definition of relations but also on apects of structure, i.e., whether a tree is sufficient as a representational device or general graphs are required (and if so, whether any restrictions can be placed on these graph\u00e2\u0080\u0099s structures \u00e2\u0080\u0094 cf.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Ex: Mr. Bush disclosed the policy by reading it...","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"95 B a s e li n e ( S e lf t a g ) 70 a l l B i k e l ( v 1 . 2 ) B a s e l i n e ( P r e t a g ) 7 0 a l l G o l d P O S 70 0.7 70 0.801 278 0.7 52 0.794 278 0.7 71 0.804 295 0.7 52 0.796 295 0.7 75 0.808 309 77.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"A similar structure is used in speech recognition.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Each name class is subdivided into 4 sub-classes, i.e., N begin, N continue, N end, and N unique.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Previous reports on Chinese segmentation have invariably cited performance either in terms of a single percent-correct score, or else a single precision-recall pair.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"There is a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099 of ten commentaries, for which the range of information (except for syntax) has been completed; the remaining data has been annotated to different degrees, as explained below.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Our work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","label":1},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"This feature includes named entities, which the ATB marks with a flat NP node dominating an arbitrary number of NNP pre-terminal daughters (Figure 2).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For a word whose initCaps might be due to its position rather than its meaning (in headlines, first word of a sentence, etc), the case information of other occurrences might be more accurate than its own.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In addition, each feature function is a binary function.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"A different notion of information structure, is used in work such as that of (?), who tried to characterize felicitous constituent ordering (theme choice, in particular) that leads to texts presenting information in a natural, \u00e2\u0080\u009cflowing\u00e2\u0080\u009d way rather than with abrupt shifts of attention.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"This scenario is applicable to a large set of languages and has been considered by a number of authors in the past (Alshawi et al., 2000; Xi and Hwa, 2005; Ganchev et al., 2009).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Another approach to finding paraphrases is to find phrases which take similar subjects and objects in large corpora by using mutual information of word distribution [Lin and Pantel 01].","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"2 for the accuracy of the different methods.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"BABAR employs information extraction techniques to represent and learn role relationships.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"68 96.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"In general, the neighborhoods can be more diverse and we allow a soft label distribution over the vertices.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Figure 1 shows sample sentences from these domains, which are widely divergent.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The number of ATB n-grams also falls below the WSJ sample size as the largest WSJ sample appeared in only 162 corpus positions.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Furthermore, we know one of the original parses will be the hypothesized parse, so the direct method of determining which one is best is to compute the probability of each of the candidate parses using the probabilistic model we developed in Section 2.1.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Reference resolution involves finding words that co-refer to the same entity.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Section 2.2 then describes our representation for contextual roles and four types of contextual role knowledge that are learned from the training examples.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Lemma: If the number of votes required by constituent voting is greater than half of the parsers under consideration the resulting structure has no crossing constituents.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We consider properties of the tree sets generated by CFG's, Tree Adjoining Grammars (TAG's), Head Grammars (HG's), Categorial Grammars (CG's), and IG's.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This is in general very difficult, given the extremely free manner in which Chinese given names are formed, and given that in these cases we lack even a family name to give the model confidence that it is identifying a name.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"36 79.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"After the first step towards breadth had been taken with the PoS-tagging, RST annotation, and URML conversion of the entire corpus of 170 texts12 , emphasis shifted towards depth.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Another way to interpret this is that less than 5% of the correct constituents are missing from the hypotheses generated by the union of the three parsers.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The 13 positions of the source sentence are processed in the order shown.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(1991}, Gu and Mao (1994), and Nie, Jin, and Hannan (1994).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The 4th block contains instance-weighting models trained on all features, used within a MAP TM combination, and with a linear LM mixture.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"As is standard, we report the greedy one-to-one (Haghighi and Klein, 2006) and the many-to-one token-level accuracy obtained from mapping model states to gold POS tags.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"SRILM (Stolcke, 2002) is widely used within academia.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"When translating the sentence monotonically from left to right, the translation of the German finite verb 'kann', which is the left verbal brace in this case, is postponed until the German noun phrase 'mein Kollege' is translated, which is the subject of the sentence.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The Verbmobil task is an appointment scheduling task.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"A minimal requirement for building a Chinese word segmenter is obviously a dictionary; furthermore, as has been argued persuasively by Fung and Wu (1994), one will perform much better at segmenting text by using a dictionary constructed with text of the same genre as the text to be segmented.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"If e < b then the key is not found.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We refer to different readings as different analyses whereby the segments are deterministic given the sequence of PoS tags.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","label":1},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"For example, one of the ATB samples was the determiner -\"\" ; dhalik\u00e2\u0080\u009cthat.\u00e2\u0080\u009d The sample occurred in 1507 corpus po sitions, and we found that the annotations were consistent.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Roughly speaking, the new algorithm presented in this paper performs a similar search, but instead minimizes a bound on the number of (unlabeled) examples on which two classifiers disagree.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"\u00ce\u00b8 has a belief value of 1.0, indicating complete certainty that the correct hypothesis is included in the set, and a plausibility value of 1.0, indicating that there is no evidence for competing hypotheses.5 As evidence is collected and the likely hypotheses are whittled down, belief is redistributed to subsets of \u00ce\u00b8.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"However, it is almost universally the case that no clear definition of what constitutes a \"correct\" segmentation is given, so these performance measures are hard to evaluate.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Evalb, the standard parsing metric, is biased toward such corpora (Sampson and Babarczy, 2003).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"In Table 1 we see with very few exceptions that the isolated constituent precision is less than 0.5 when we use the constituent label as a feature.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"For a corpus C, let M be the set of tuples \u00e2\u0088\u0097n, l), where n is an n-gram with bracketing label l. If any n appears 6 Generative parsing performance is known to deteriorate with sentence length.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The distribution specifies the relative weight, or importance, of each example \u2014 typically, the weak learner will attempt to minimize the weighted error on the training set, where the distribution specifies the weights.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"ICOC and CSPP contributed the greatest im provements.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This can be repeated several times to collect a list of author \/ book title pairs and expressions.","label":0},{"source_text":"Here we present two algorithms.","target_text":"For the purposes of EM, the &quot;observed&quot; data is {(xi, Ya\u2022 \u2022 \u2022 (xrn, Yrn), xfil, and the hidden data is {ym+i y}.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"(S; C; j); Not only the coverage set C and the positions j; j0, but also the verbgroup states S; S0 are taken into account.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, there will remain a large number of words that are not readily adduced to any produc\u00c2\u00ad tive pattern and that would simply have to be added to the dictionary.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The following auxiliary quantity is defined: Qe0 (e; C; j) := probability of the best partial hypothesis (ei 1; bi 1), where C = fbkjk = 1; ; ig, bi = j, ei = e and ei\u00f4\u0080\u0080\u00801 = e0.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"We can do that . Input: Das ist zu knapp , weil ich ab dem dritten in Kaiserslautern bin . Genaugenommen nur am dritten . Wie ware es denn am ahm Samstag , dem zehnten Februar ? MonS: That is too tight , because I from the third in Kaiserslautern . In fact only on the third . How about ahm Saturday , the tenth of February ? QmS: That is too tight , because I am from the third in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? IbmS: That is too tight , from the third because I will be in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? Input: Wenn Sie dann noch den siebzehnten konnten , ware das toll , ja . MonS: If you then also the seventeenth could , would be the great , yes . QmS: If you could then also the seventeenth , that would be great , yes . IbmS: Then if you could even take seventeenth , that would be great , yes . Input: Ja , das kommt mir sehr gelegen . Machen wir es dann am besten so . MonS: Yes , that suits me perfectly . Do we should best like that . QmS: Yes , that suits me fine . We do it like that then best . IbmS: Yes , that suits me fine . We should best do it like that .","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Table 2 shows these similarity measures.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Fortunately, performance was stable across various values, and we were able to use the same hyperparameters for all languages.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Examples will usually be accompanied by a translation, plus a morpheme-by-morpheme gloss given in parentheses whenever the translation does not adequately serve this purpose.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, co-occurring caseframes may reflect synonymy (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> abducted\u00e2\u0080\u009d) or related events (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> released\u00e2\u0080\u009d).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Using these states, we define partial hypothesis extensions, which are of the following type: (S0;C n fjg; j0) !","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Each unlabeled pair (x1,i, x2,i) is represented as an edge between nodes corresponding to x1,i and X2,i in the graph.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"How should the absence of vowels and syntactic markers influence annotation choices and grammar development?","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"2.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In addition, there are several approaches to non-projective dependency parsing that are still to be evaluated in the large (Covington, 1990; Kahane et al., 1998; Duchier and Debusmann, 2001; Holan et al., 2001; Hellwig, 2003).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Our annotators pointed out that very often they made almost random decisions as to what relation to choose, and where to locate the boundary of a span.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Keys to the table are hashed, using for example Austin Appleby\u2019s MurmurHash2, to integers evenly distributed over a large range.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Each distance in the traveling salesman problem now corresponds to the negative logarithm of the product of the translation, alignment and language model probabilities.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Second, we identified the 100 most frequent nouns in the training corpus and manually labeled them with semantic tags.","label":0},{"source_text":"Here we present two algorithms.","target_text":"As in boosting, the algorithm works in rounds.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Annotation consistency is important in any supervised learning task.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"We will briefly discuss this point in Section 3.1.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"These performance gains transfer to improved system runtime performance; though we focused on Moses, our code is the best lossless option with cdec and Joshua.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"2.4 Underspecified rhetorical structure.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The type-level posterior term can be computed according to, P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2) \u00e2\u0088\u009d Note that each round of sampling Ti variables takes time proportional to the size of the corpus, as with the standard token-level HMM.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The corresponding token words w are drawn conditioned on t and \u00ce\u00b8.2 Our full generative model is given by: K P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) = n (P (\u00cf\u0086t|\u00ce\u00b1)P (\u00ce\u00b8t|T , \u00ce\u00b1)) t=1 The transition distribution \u00cf\u0086t for each tag t is drawn according to DIRICHLET(\u00ce\u00b1, K ), where \u00ce\u00b1 is the shared transition and emission distribution hyperparameter.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We used 22 features for the logistic weighting model, divided into two groups: one intended to reflect the degree to which a phrase pair belongs to general language, and one intended to capture similarity to the IN domain.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This is motivated by taking \u03b2 po(s|t) to be the parameters of a Dirichlet prior on phrase probabilities, then maximizing posterior estimates p(s|t) given the IN corpus.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"A better approach would be to distin guish between these cases, possibly by drawing on the vast linguistic work on Arabic connectives (AlBatal, 1990).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"att.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"10 Other orthographic normalization schemes have been suggested for Arabic (Habash and Sadat, 2006), but we observe negligible parsing performance differences between these and the simple scheme used in this evaluation.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"mein 5.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"irL as the product of the probability estimate for i\u00c2\u00a5JJ1l., and the probability estimate just derived for unseen plurals in ir,: p(i\u00c2\u00a51J1l.ir,) p(i\u00c2\u00a51J1l.)p(unseen(f,)).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We run the baseline Moses system for the French-English track of the 2011 Workshop on Machine Translation,9 translating the 3003-sentence test set.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Ex: The government said it ...","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Recent research has demonstrated that incorporating this sparsity constraint improves tagging accuracy.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In general, m; l; l0 6= fl1; l2; l3g and in line umber 3 and 4, l0 must be chosen not to violate the above reordering restriction.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In fact, during the first rounds many of the predictions of Th., g2 are zero.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Therefore, we only score guess\/gold pairs with identical character yields, a condition that allows us to measure parsing, tagging, and segmentation accuracy by ignoring whitespace.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The PROBING model was designed to improve upon SRILM by using linear probing hash tables (though not arranged in a trie), allocating memory all at once (eliminating the need for full pointers), and being easy to compile.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"I = 1X21 N and N is a &quot;medium&quot; sized number so that it is feasible to collect 0(N) unlabeled examples.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The number of ATB n-grams also falls below the WSJ sample size as the largest WSJ sample appeared in only 162 corpus positions.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Mikheev et al.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We compared the ATB5 to tree- banks for Chinese (CTB6), German (Negra), and English (WSJ) (Table 4).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"This algorithm can be applied to statistical machine translation.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"There is a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099 of ten commentaries, for which the range of information (except for syntax) has been completed; the remaining data has been annotated to different degrees, as explained below.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"G1 and G2 are hanzi, we can estimate the probability of the sequence being a name as the product of: \u00e2\u0080\u00a2 the probability that a word chosen randomly from a text will be a name-p(rule 1), and \u00e2\u0080\u00a2 the probability that the name is of the form 1hanzi-family 2hanzi-given-p(rule 2), and \u00e2\u0080\u00a2 the probability that the family name is the particular hanzi F1-p(rule 6), and \u00e2\u0080\u00a2 the probability that the given name consists of the particular hanzi G1 and G2-p(rule 9) This model is essentially the one proposed in Chang et al.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Then, moving from connective to connective, ConAno sometimes offers suggestions for its scope (using heuristics like \u00e2\u0080\u0098for sub- junctor, mark all words up to the next comma as the first segment\u00e2\u0080\u0099), which the annotator can accept with a mouseclick or overwrite, marking instead the correct scope with the mouse.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In fact, training on the official training data is not suitable as the articles in this data set are entirely about aviation disasters, and the test data is about air vehicle launching.","label":0},{"source_text":"There is no global pruning.","target_text":"A detailed description of the search procedure used is given in this patent.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Lexicon and OOV Handling Our data-driven morphological-analyzer proposes analyses for unknown tokens as described in Section 5.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR uses a named entity recognizer to identify proper names that refer to people and companies.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"2.4 Underspecified rhetorical structure.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"If there is a frequent multi-word sequence in a domain, we could use it as a keyword candidate.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"(2010) consistently outperforms ours on English, we obtain substantial gains across other languages.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"DempsterShafer handles this by re-normalizing all the belief values with respect to only the non-null sets (this is the purpose of the denominator in Equation 1).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"A graphical depiction of our model as well as a summary of random variables and parameters can be found in Figure 1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Precision.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"In both cases, SRILM walks its trie an additional time to minimize context as mentioned in Section 4.1.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We picked two domains, the CC-domain and the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain (PC-domain), for the evaluation, as the entire system output was too large to evaluate.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"It reads a file with a list of German connectives, and when a text is opened for annotation, it highlights all the words that show up in this list; these will be all the potential connectives.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"1).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"And indeed, converging on annotation guidelines is even more difficult than it is with co-reference.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(2003), which gives 96.8% accuracy on the test set.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Note, however, that there might be situations in which Zco in fact increases.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: Mr. Cristiani is the president ...","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(1998) have also used a maximum entropy classifier that uses already tagged entities to help tag other entities.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Given an anaphor, BABAR identifies the caseframe that would extract it from its sentence.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Here a set is represented by the keyword and the number in parentheses indicates the number of shared NE pair instances.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"As we have said, parse quality decreases with sentence length.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"However, this optimistic search would not visit the entries necessary to store backoff information in the outgoing state.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In fact, training on the official training data is not suitable as the articles in this data set are entirely about aviation disasters, and the test data is about air vehicle launching.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Hence, for each token, one of the four features zone-TXT, zone- HL, zone-DATELINE, or zone-DD is set to 1, and the other 3 are set to 0.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"6 Joint Segmentation and Parsing.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The tool we use is MMAX8, which has been specifically designed for marking co-reference.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Figure 2 shows an excerpt of a sentence from the Italian test set and the tags assigned by four different models, as well as the gold tags.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"0 D\/ nc 5.0 The minimal dictionary encoding this information is represented by the WFST in Figure 2(a).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In this domain the major scenarios involve the things they agreed on, rather than the mere fact that they agreed.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"During training, we treat as observed the language word types W as well as the token-level corpus w. We utilize Gibbs sampling to approximate our collapsed model posterior: P (T ,t|W , w, \u00ce\u00b1, \u00ce\u00b2) \u00e2\u0088\u009d P (T , t, W , w|\u00ce\u00b1, \u00ce\u00b2) 0.7 0.6 0.5 0.4 0.3 English Danish Dutch Germany Portuguese Spanish Swedish = P (T , t, W , w, \u00cf\u0088, \u00ce\u00b8, \u00cf\u0086, w|\u00ce\u00b1, \u00ce\u00b2)d\u00cf\u0088d\u00ce\u00b8d\u00cf\u0086 Note that given tag assignments T , there is only one setting of token-level tags t which has mass in the above posterior.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Additionally, for a given coverage set, at most 250 different hypotheses are kept during the search process, and the number of different words to be hypothesized by a source word is limited.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The possible analyses of a surface token pose constraints on the analyses of specific segments.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A detailed description of the search procedure used is given in this patent.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"In this section, we extend state to optimize left-to-right queries.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"To define a similarity function between the English and the foreign vertices, we rely on high-confidence word alignments.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(1) CEO of McCann . . .","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The corpus-based statistical parsing community has many fast and accurate automated parsing systems, including systems produced by Collins (1997), Charniak (1997) and Ratnaparkhi (1997).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Hence we decided to restrict ourselves to only information from the same document.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"For example, Gazdar (1985) discusses the applicability of Indexed Grammars (IG's) to Natural Language in terms of the structural descriptions assigned; and Berwick (1984) discusses the strong generative capacity of Lexical-Functional Grammar (LFG) and Government and Bindings grammars (GB).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"He also argues that paired English complementizers may also require structural descriptions whose path sets have nested dependencies.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"AdaBoost was first introduced in (Freund and Schapire 97); (Schapire and Singer 98) gave a generalization of AdaBoost which we will use in this paper.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Though we are not able to calculate their memory usage on our model, results reported in their paper suggest lower memory consumption than TRIE on large-scale models, at the expense of CPU time.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"com t 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"These alignment models are similar to the concept of hidden Markov models (HMM) in speech recognition.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Maamouri et al.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We create equivalence classes for verb, noun, and adjective POS categories.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"25 16.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Once HMM parameters (\u00ce\u00b8, \u00cf\u0086) are drawn, a token-level tag and word sequence, (t, w), is generated in the standard HMM fashion: a tag sequence t is generated from \u00cf\u0086.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"In Figure 4 we show an example of variation between the parsing models.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Step 2.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The translation search is carried out with the category markers and the city names are resubstituted into the target sentence as a postprocessing step.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Since the transducers are built from human-readable descriptions using a lexical toolkit (Sproat 1995), the system is easily maintained and extended.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The developers suggested some changes, such as building the model from scratch with IRSTLM, but these did not resolve the problem.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"It also uses less memory, with 8 bytes of overhead per entry (we store 16-byte entries with m = 1.5); linked list implementations hash set and unordered require at least 8 bytes per entry for pointers.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In our situation, the competing hypotheses are the possible antecedents for an anaphor.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"May 1995).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The combining technique must act as a multi-position switch indicating which parser should be trusted for the particular sentence.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A Stochastic Finite-State Word-Segmentation Algorithm for Chinese","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"In line with perplexity results from Table 1, the PROBING model is the fastest followed by TRIE, and subsequently other packages.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"For effectively annotating connectives\/scopes, we found that existing annotation tools were not well-suited, for two reasons: \u00e2\u0080\u00a2 Some tools are dedicated to modes of annotation (e.g., tiers), which could only quite un-intuitively be used for connectives and scopes.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Despite these limitations, a purely finite-state approach to Chinese word segmentation enjoys a number of strong advantages.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"4.1 Local Features.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Analogous to the sharing of stacks in IC's, Lexical-Functional Grammar's (LFG's) use the unification of unbounded hierarchical structures.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"In general, several gross corpus statistics favor the ATB, so other factors must contribute to parsing underperformance.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"5 67.3 55.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"6 Our knowledge sources return some sort of probability estimate, although in some cases this estimate is not especially well-principled (e.g., the Recency KS).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"A greedy algorithm (or maximum-matching algorithm), GR: proceed through the sentence, taking the longest match with a dictionary entry at each point.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Model components cascade, so the row corresponding to +FEATS also includes the PRIOR component (see Section 3).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Cur In order to ground such approaches in linguistic observation and description, a multi-level anno 10 For an exposition of the idea as applied to the task of text planning, see (Chiarcos, Stede 2004).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Previous approaches have tried to find examples that are similar to the target domain.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We incorporate instance weighting into a mixture-model framework, and find that it yields consistent improvements over a wide range of baselines.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Simply assigning to each word its most frequent associated tag in a corpus achieves 94.6% accuracy on the WSJ portion of the Penn Treebank.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"These packages are further described in Section 3.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"level.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Otherwise for the predecessor search hypothesis, we would have chosen a position that would not have been among the first n uncovered positions.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"63 81.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"A named entity recognizer (NER) is useful in many NLP applications such as information extraction, question answering, etc. On its own, a NER can also provide users who are looking for person or organization names with quick information.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Semantic expectations are analogous to lexical expectations except that they represent semantic classes rather than nouns.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Let n be some node labeled X in a tree -y (see Figure 3).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(3)In sentence (1), McCann can be a person or an orga nization.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The pre terminal morphological analyses are mapped to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags provided with the tree- bank.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"DempsterShafer handles this by re-normalizing all the belief values with respect to only the non-null sets (this is the purpose of the denominator in Equation 1).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Each name class is subdivided into 4 sub-classes, i.e., N begin, N continue, N end, and N unique.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The system of Berg-Kirkpatrick et al.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"If enough parsers suggest that a particular constituent belongs in the parse, we include it.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"One side of the decision making process is when we choose to believe a constituent should be in the parse, even though only one parser suggests it.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We adapt the string pumping lemma for the class of languages corresponding to the complexity of the path set.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"On each step CoBoost searches for a feature and a weight so as to minimize either 40 or 40.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(3)), with one term for each classifier.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For example, in the phrase \u00e2\u0080\u009cCompany-A last week purchased rival Marshalls from Company-B\u00e2\u0080\u009d, the purchased company is Marshalls, not Company-B.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"i=1 (f,v)\u00e2\u0088\u0088Wi","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We use Alternating Turing Machines (Chandra, Kozen, and Stockmeyer, 1981) to show that polynomial time recognition is possible for the languages discussed in Section 4.3.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Finally, a DempsterShafer probabilistic model evaluates the evidence provided by the knowledge sources for all candidate antecedents and makes the final resolution decision.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Microsoft\u2019s approach uses dependency trees, others use hierarchical phrase models.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"annotation guidelines that tell annotators what to do in case of doubt.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The string pumping lemma for CFG's (uvwxy-theorem) can be seen as a corollary of this lemma. from this pumping lemma: a single path can be pumped independently.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The final strong hypothesis, denoted 1(x), is then the sign of a weighted sum of the weak hypotheses, 1(x) = sign (Vii atht(x)), where the weights at are determined during the run of the algorithm, as we describe below.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In Input: {(x1,i, Initialize: Vi, j : e(xi) = 0.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"That is, we can use the discourse parser on PCC texts, emulating for instance a \u00e2\u0080\u009cco-reference oracle\u00e2\u0080\u009d that adds the information from our co-reference annotations.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Several methods exist to handle collisions; we use linear probing because it has less memory overhead when entries are small.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"For this experiment, we compare our model with the uniform tag assignment prior (1TW) with the learned prior (+PRIOR).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"What ought to be developed now is an annotation tool that can make use of the format, allow for underspecified annotations and visualize them accordingly.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Such constraints are derived from training data, expressing some relationship between features and outcome.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Still, from a theoretical point of view, projective parsing of non-projective structures has the drawback that it rules out perfect accuracy even as an asymptotic goal.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"For graph propagation, the hyperparameter v was set to 2 x 10\u22126 and was not tuned.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"of Tokens No.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Several coreference resolvers have used supervised learning techniques, such as decision trees and rule learners (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The portions of information in the large window can be individually clicked visible or invisible; here we have chosen to see (from top to bottom) \u00e2\u0080\u00a2 the full text, \u00e2\u0080\u00a2 the annotation values for the activated annotation set (co-reference), \u00e2\u0080\u00a2 the actual annotation tiers, and \u00e2\u0080\u00a2 the portion of text currently \u00e2\u0080\u0098in focus\u00e2\u0080\u0099 (which also appears underlined in the full text).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The general-language features have a slight advantage over the similarity features, and both are better than the SVM feature.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In Section 3, we introduce our novel concept to word reordering and a DP-based search, which is especially suitable for the translation direction from German to English.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Also, expanding on the techniques for the automatic generation of extraction patterns (Riloff 96; Sudo 03) using our method, the extraction patterns which have the same meaning can be automatically linked, enabling us to produce the final table fully automatically.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Their results are then compared with the results of an automatic segmenter.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We use label propagation in two stages to generate soft labels on all the vertices in the graph.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Word Re-ordering and DP-based Search in Statistical Machine Translation","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"For each language, we took the same number of sentences from the bitext as there are in its treebank, and trained a supervised feature-HMM.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Along with locking and background kernel operations such as prefaulting, this explains why wall time is not one-eighth that of the single-threaded case. aLossy compression with the same weights. bLossy compression with retuned weights. the non-lossy options.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"To date we have not done a separate evaluation of foreign-name recognition.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Email: cls@bell-labs.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"x EG, has agreed to be bought by H x EG, now owned by H x H to acquire EG x H\u00e2\u0080\u0099s agreement to buy EG Three of those phrases are actually paraphrases, but sometime there could be some noise; such as the second phrase above.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Examples are given in Table 4.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The BerkeleyLM direct-mapped cache is in principle faster than caches implemented by RandLM and by IRSTLM, so we may write a C++ equivalent implementation as future work.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"For this experiment, we compare our model with the uniform tag assignment prior (1TW) with the learned prior (+PRIOR).","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Quite often, though, these directives fulfill the goal of increasing annotator agreement without in fact settling the theoretical question; i.e., the directives are clear but not always very well motivated.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"All the NE pair instances which co-occur separated by at most 4 chunks are collected along with information about their NE types and the phrase between the NEs (the \u00e2\u0080\u0098context\u00e2\u0080\u0099).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In many cases, there is an even stronger restriction: over large portions of the source string, the alignment is monotone.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Sorted arrays store key-value pairs in an array sorted by key, incurring no space overhead.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Virginia) and -sia are normally transliterated as fbSi!","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Human evaluation is one way to distinguish between the two cases.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The first step in the learning process is to generate training examples consisting of anaphor\/antecedent resolutions.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"(2010), we adopt a simpler na\u00c2\u00a8\u00c4\u00b1ve Bayes strategy, where all features are emitted independently.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"On the other hand, we are interested in the application of rhetorical analysis or \u00e2\u0080\u0098discourse parsing\u00e2\u0080\u0099 (3.2 and 3.3), in text generation (3.4), and in exploiting the corpus for the development of improved models of discourse structure (3.5).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Training examples are generated automatically by identifying noun phrases that can be easily resolved with their antecedents using lexical and syntactic heuristics.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"3 The Coreference Resolution Model.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Then we ran binary search to determine the least amount of memory with which it would run.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In previous work (Bean and Riloff, 1999), we developed an unsupervised learning algorithm that automatically recognizes definite NPs that are existential without syntactic modification because their meaning is universally understood.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"17 They also provide a set of title-driven rules to identify names when they occur before titles such as $t. 1: xianlshengl 'Mr.' or i:l:itr!J tai2bei3 shi4zhang3 'Taipei Mayor.'","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"For the experiments, we used four newswire corpora, the Los Angeles Times\/Washington Post, The New York Times, Reuters and the Wall Street Journal, all published in 1995.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"While building a machine translation system is a serious undertaking, in future we hope to attract more newcomers to the field by keeping the barrier of entry as low as possible.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"However, MADA is language-specific and relies on manually constructed dictionaries.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"so that 'door' would be and in this case the hanzi 7C, does not represent a syllable.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"For English POS tagging, BergKirkpatrick et al. (2010) found that this direct gradient method performed better (>7% absolute accuracy) than using a feature-enhanced modification of the Expectation-Maximization (EM) algorithm (Dempster et al., 1977).8 Moreover, this route of optimization outperformed a vanilla HMM trained with EM by 12%.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We also report word type level accuracy, the fraction of word types assigned their majority tag (where the mapping between model state and tag is determined by greedy one-to-one mapping discussed above).5 For each language, we aggregate results in the following way: First, for each hyperparameter setting, evaluate three variants: The first model (1TW) only 4 Typically, the performance stabilizes after only 10 itera-.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"It also uses less memory, with 8 bytes of overhead per entry (we store 16-byte entries with m = 1.5); linked list implementations hash set and unordered require at least 8 bytes per entry for pointers.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Also, \u00e2\u0080\u009cagree\u00e2\u0080\u009d in the CC-domain is not a desirable keyword.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"It has been shown for English (Wang and Hirschberg 1992; Hirschberg 1993; Sproat 1994, inter alia) that grammatical part of speech provides useful information for these tasks.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Evaluation results within sets Table 1 shows the evaluation result based on the number of phrases in a set.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Since foreign names can be of any length, and since their original pronunciation is effectively unlimited, the identi\u00c2\u00ad fication of such names is tricky.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Therefore we opted to use the much 7 The DempsterShafer theory assumes that one of the hypotheses in \u00ce\u00b8 is correct, so eliminating all of the hypotheses violates this assumption.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"input: source string f1:::fj :::fJ initialization for each cardinality c = 1; 2; ; J do for each pair (C; j), where j 2 C and jCj = c do for each target word e 2 E Qe0 (e; C; j) = p(fj je) max \u00c3\u0086;e00 j02Cnfjg fp(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j0)g words fj in the input string of length J. For the final translation each source position is considered exactly once.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The discovered paraphrases can be a big help to reduce human labor and create a more comprehensive pattern set.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"na me =>2 ha nzi fa mi ly 2 ha nzi gi ve n 5.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Finally, we find links between sets of phrases, based on the NE instance pair data (for example, different phrases which link \u00e2\u0080\u009cIBM\u00e2\u0080\u009d and \u00e2\u0080\u009cLotus\u00e2\u0080\u009d) (Step 4).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"A high-level relation is agent, which relates an animate nominal to a predicate.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Morphologically derived words such as, xue2shengl+men0.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Context from the whole document can be important in classifying a named entity.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The DempsterShafer rule for combining pdfs is: to {C}, meaning that it is 70% sure the correct hypothesis is C. The intersection of these sets is the null set because these beliefs are contradictory.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Like verbs, maSdar takes arguments and assigns case to its objects, whereas it also demonstrates nominal characteristics by, e.g., taking determiners and heading iDafa (Fassi Fehri, 1993).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"(we ignored the 419 trees in their development set.)","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Here a set is represented by the keyword and the number in parentheses indicates the number of shared NE pair instances.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Next, we represent the input sentence as an unweighted finite-state acceptor (FSA) I over H. Let us assume the existence of a function Id, which takes as input an FSA A, and produces as output a transducer that maps all and only the strings of symbols accepted by A to themselves (Kaplan and Kay 1994).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The NIST06 and NIST08 evaluation sets were used for testing.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Simply assigning to each word its most frequent associated tag in a corpus achieves 94.6% accuracy on the WSJ portion of the Penn Treebank.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"However, TRIE partitions storage by n-gram length, so walking the trie reads N disjoint pages.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Besides the lack of a clear definition of what constitutes a correct segmentation for a given Chinese sentence, there is the more general issue that the test corpora used in these evaluations differ from system to system, so meaningful comparison between systems is rendered even more difficult.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This means that the rules in our grammar are of two kinds: (a) syntactic rules relating nonterminals to a sequence of non-terminals and\/or PoS tags, and (b) lexical rules relating PoS tags to lattice arcs (lexemes).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"On each step CoBoost searches for a feature and a weight so as to minimize either 40 or 40.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We used this data to build an unpruned ARPA file with IRSTLM\u2019s improved-kneser-ney option and the default three pieces.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Unlike the WSJ corpus which has a high frequency of rules like VP \u00e2\u0086\u0092VB PP, Arabic verb phrases usually have lexi calized intervening nodes (e.g., NP subjects and direct objects).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The first author is supported by a National Defense Science and Engineering Graduate (NDSEG) fellowship.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"1 is given in Fig.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Case and Zone of and : Similarly, if (or ) is initCaps, a feature (initCaps, zone) (or (initCaps, zone) ) is set to 1, etc. Token Information: This group consists of 10 features based on the string , as listed in Table 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In the following, we assume that this word joining has been carried out.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The distribution specifies the relative weight, or importance, of each example \u2014 typically, the weak learner will attempt to minimize the weighted error on the training set, where the distribution specifies the weights.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each candidate antecedent, BABAR identifies the caseframe that would extract the candidate, pairs it with the anaphor\u00e2\u0080\u0099s caseframe, and consults the CF Network to see if this pair of caseframes has co-occurred in previous resolutions.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"5 67.3 55.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We would like to thank Eugene Charniak, Michael Collins, and Adwait Ratnaparkhi for enabling all of this research by providing us with their parsers and helpful comments.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, this result is consistent with the results of ex\u00c2\u00ad periments discussed in Wu and Fung (1994).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Each round is composed of two stages; each stage updates one of the classifiers while keeping the other classifier fixed.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Chang of Tsinghua University, Taiwan, R.O.C., for kindly providing us with the name corpora.","label":0},{"source_text":"There is no global pruning.","target_text":"Each distance in the traveling salesman problem now corresponds to the negative logarithm of the product of the translation, alignment and language model probabilities.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, it is almost universally the case that no clear definition of what constitutes a \"correct\" segmentation is given, so these performance measures are hard to evaluate.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We use double-circles to indicate the space-delimited token boundaries.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"It was motivated by the observation that the (Yarowsky 95) algorithm added a very large number of rules in the first few iterations.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The first shows how constituent features and context do not help in deciding which parser to trust.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Sc opi ng filters candidate if outside the anaphor\u00e2\u0080\u0099s scope.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Here, \u00e2\u0080\u009cEG\u00e2\u0080\u009d represents \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"If we were working with more than three parsers we could investigate minority constituents, those constituents that are suggested by at least one parser, but which the majority of the parsers do not suggest.","label":0},{"source_text":"There is no global pruning.","target_text":"Input: Ja , wunderbar . Konnen wir machen . MonS: Yes, wonderful.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Second, we identified the 100 most frequent nouns in the training corpus and manually labeled them with semantic tags.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"If a candidate has a belief value \u00e2\u0089\u00a5 .50, then we select that candidate as the antecedent for the anaphor.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We include a constituent in our hypothesized parse if it appears in the output of a majority of the parsers.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For E(ni1s), then, we substitute a smooth S against the number of class elements.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"There are two possible reasons: (1) the knowledge sources are resolving different cases of anaphora, and (2) the knowledge sources provide multiple pieces of evidence in support of (or against) a candidate, thereby acting synergistically to push the DempsterShafer model over the belief threshold in favor of a single candidate.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"A dynamic programming algorithm is then used to select the sequence of word classes with the highest probability.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We reduce this to O(log log |A|) time by evenly distributing keys over their range then using interpolation search4 (Perl et al., 1978).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"As long as the main evaluation metric is dependency accuracy per word, with state-of-the-art accuracy mostly below 90%, the penalty for not handling non-projective constructions is almost negligible.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"With the same Corporate- Suffix-List and Person-Prefix-List used in local features, for a token seen elsewhere in the same document with one of these suffixes (or prefixes), another feature Other-CS (or Other-PP) is set to 1.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"In the rhetorical tree, nuclearity information is then used to extract a \u00e2\u0080\u009ckernel tree\u00e2\u0080\u009d that supposedly represents the key information from which the summary can be generated (which in turn may involve co-reference information, as we want to avoid dangling pronouns in a summary).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The weak hypothesis can abstain from predicting the label of an instance x by setting h(x) = 0.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Not all the layers have been produced for all the texts yet.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Here a set is represented by the keyword and the number in parentheses indicates the number of shared NE pair instances.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In many cases, there is an even stronger restriction: over large portions of the source string, the alignment is monotone.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This decreases the statistical significance of our results compared to those studies.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"More details on the parsing algorithm can be found in Nivre (2003).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"x EG, has agreed to be bought by H x EG, now owned by H x H to acquire EG x H\u00e2\u0080\u0099s agreement to buy EG Three of those phrases are actually paraphrases, but sometime there could be some noise; such as the second phrase above.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Furthermore, some of the arcs represent lexemes not present in the input tokens (e.g. h\/DT, fl\/POS), however these are parts of valid analyses of the token (cf. super-segmental morphology section 2).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Suffixes and Prefixes: This group contains only two features: Corporate-Suffix and Person-Prefix.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Since these are distinct syntactic units, they are typically segmented.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"For each terminal, the Leaf Ancestor metric extracts the shortest path to the root.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"From here on we will refer to the named-entity string itself as the spelling of the entity, and the contextual predicate as the context.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Our use of an unweighted lattice reflects our belief that all the segmentations of the given input sentence are a-priori equally likely; the only reason to prefer one segmentation over the another is due to the overall syntactic context which is modeled via the PCFG derivations.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Statistics are printed before Moses exits and after parts of the decoder have been destroyed.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In our experiments we set the parameter values randomly, and then ran EM to convergence.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Out-of-Vocabulary: We derived a lexicon list from WordNet 1.6, and words that are not found in this list have a feature out-of-vocabulary set to 1.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We substantially outperform all of them on query speed and offer lower memory consumption than lossless alternatives.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Using the concept of inverted alignments, we explicitly take care of the coverage constraint by introducing a coverage set C of source sentence positions that have been already processed.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The kind of annotation work presented here would clearly benefit from the emergence of standard formats and tag sets, which could lead to sharable resources of larger size.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For all other recursive NPs, we add a common annotation to the POS tag of the head (recursiveNPHead).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"In all of our experiments, the binary file (whether mapped or, in the case of most other packages, interpreted) is loaded into the disk cache in advance so that lazy mapping will never fault to disk.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"na me =>2 ha nzi fa mi ly 1 ha nzi gi ve n 6.1 ha nzi fa mi ly => ha nz ii 7.2 ha nzi fa mi ly => ha nzi i ha nz ij 8.1 ha nzi gi ve n => ha nz ii 9.2 ha nzi giv en => ha nzi i ha nz ij The difficulty is that given names can consist, in principle, of any hanzi or pair of hanzi, so the possible given names are limited only by the total number of hanzi, though some hanzi are certainly far more likely than others.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"All three models evaluated in this paper incorrectly analyze the constituent as iDafa; none of the models attach the attributive adjectives properly.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The first probability is estimated from a name count in a text database, and the rest of the probabilities are estimated from a large list of personal names.n Note that in Chang et al.'s model the p(rule 9) is estimated as the product of the probability of finding G 1 in the first position of a two-hanzi given name and the probability of finding G2 in the second position of a two-hanzi given name, and we use essentially the same estimate here, with some modifications as described later on.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Ties are rare in Bayes switching because the models are fine-grained \u2014 many estimated probabilities are involved in each decision.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Also, in Information Extraction (IE), in which the system tries to extract elements of some events (e.g. date and company names of a corporate merger event), several event instances from different news articles have to be aligned even if these are expressed differently.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Syntactic decoders, such as cdec (Dyer et al., 2010), build state from null context then store it in the hypergraph node for later extension.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If the token is the first word of a sentence, then this feature is set to 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The monotone search performs worst in terms of both error rates mWER and SSER.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, both a chair and a truck would be labeled as artifacts, but this does not at all suggest that they are coreferent.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kind of cue.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In the pinyin transliterations a dash(-) separates syllables that may be considered part of the same phonological word; spaces are used to separate plausible phonological words; and a plus sign (+) is used, where relevant, to indicate morpheme boundaries of interest.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In this way, the method reported on here will necessarily be similar to a greedy method, though of course not identical.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In the namedentity problem each example is a (spelling,context) pair.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The authors acknowledge the support of the NSF (CAREER grant IIS0448168, and grant IIS 0904684).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The system will then look for sequences of initial capitalized words that match the acronyms found in the whole document.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"A graphical depiction of our model as well as a summary of random variables and parameters can be found in Figure 1.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Typical data structures are generalized Bloom filters that guarantee a customizable probability of returning the correct answer.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Here, the term frequency (TF) is the frequency of a word in the bag and the inverse term frequency (ITF) is the inverse of the log of the frequency in the entire corpus.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In this situation, BABAR takes the conservative approach and declines to make a resolution.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Asking the annotator to also formulate the question is a way of arriving at more reproducible decisions.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This limitation is the obstacle to making the technology \u00e2\u0080\u009copen domain\u00e2\u0080\u009d.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Intuitively, it places more weight on OUT when less evidence from IN is available.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"However, it is also mostly political content (even if not focused on the internal workings of the European Union) and opinion.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The PROBING model was designed to improve upon SRILM by using linear probing hash tables (though not arranged in a trie), allocating memory all at once (eliminating the need for full pointers), and being easy to compile.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The algorithm, called CoBoost, has the advantage of being more general than the decision-list learning alInput: (xi , yi), , (xim, ) ; x, E 2x, yi = +1 Initialize Di (i) = 1\/m.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"2) An improved language model, which takes into account syntactic structure, e.g. to ensure that a proper English verbgroup is generated.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Inspection of the data shows that at n = 2500, the two classifiers both give labels on 44,281 (49.2%) of the unlabeled examples, and give the same label on 99.25% of these cases.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"gaolgaolxing4xing4 'happily' In the particular form of A-not-A reduplication illustrated in (3a), the first syllable of the verb is copied, and the negative markerbu4 'not' is inserted between the copy and the full verb.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"These universal POS categories not only facilitate the transfer of POS information from one language to another, but also relieve us from using controversial evaluation metrics,2 by establishing a direct correspondence between the induced hidden states in the foreign language and the observed English labels.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"These tag distributions are used to initialize the label distributions over the English vertices in the graph.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Most computational models for coreference resolution rely on properties of the anaphor and candidate antecedent, such as lexical matching, grammatical and syntactic features, semantic agreement, and positional information.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This is a rather important source of errors in name identifi\u00c2\u00ad cation, and it is not really possible to objectively evaluate a name recognition system without considering the main lexicon with which it is used.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"During development, we sensed that the Recency and Syn- role KSs did not deserve to be on equal footing with the other KSs because their knowledge was so general.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Tokens were converted to vocabulary identifiers in advance and state was carried from each query to the next.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We measured recall (Rec), precision (Pr), and the F-measure (F) with recall and precision equally weighted.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The productions of HG's are very similar to those of CFG's except that the operation used must be made explicit.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Lexicalizing several POS tags improves performance.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The second algorithm builds on a boosting algorithm called AdaBoost (Freund and Schapire 97; Schapire and Singer 98).","label":1},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"As can be seen in Figure 3, the phrases in the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set include completely different relationships, which are not paraphrases.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The test accuracy more or less asymptotes.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Using the virtual distribution Di (i) and pseudo-labels&quot;y.,\u201e values for Wo, W\u00b1 and W_ can be calculated for each possible weak hypothesis (i.e., for each feature x E Xi); the weak hypothesis with minimal value for Wo + 2\/WW _ can be chosen as before; and the weight for this weak hypothesis at = ln ww+411:) can be calculated.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Quite often, though, these directives fulfill the goal of increasing annotator agreement without in fact settling the theoretical question; i.e., the directives are clear but not always very well motivated.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We focus here instead on adapting the two most important features: the language model (LM), which estimates the probability p(wIh) of a target word w following an ngram h; and the translation models (TM) p(slt) and p(t1s), which give the probability of source phrase s translating to target phrase t, and vice versa.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"3 61.7 38.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The system will then look for sequences of initial capitalized words that match the acronyms found in the whole document.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"For example, in the phrase \u00e2\u0080\u009cCompany-A last week purchased rival Marshalls from Company-B\u00e2\u0080\u009d, the purchased company is Marshalls, not Company-B.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"including Third Tone Sandhi (Shih 1986), which changes a 3 (low) tone into a 2 (rising) tone before another 3 tone: 'j\";gil, xiao3 [lao3 shu3] 'little rat,' becomes xiao3 { lao2shu3 ], rather than xiao2 { lao2shu3 ], because the rule first applies within the word lao3shu3 'rat,' blocking its phrasal application.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Central to our approach (see Algorithm 1) is a bilingual similarity graph built from a sentence-aligned parallel corpus.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We substantially outperform all of them on query speed and offer lower memory consumption than lossless alternatives.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"12 One class of full personal names that this characterization does not cover are married women's names.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The trie data structure is commonly used for language modeling.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Dagan and Itai (Dagan and Itai, 1990) experimented with co-occurrence statistics that are similar to our lexical caseframe expectations.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The use of ILP in learning the desired grammar significantly increases the computational complexity of this method.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"First, we learn weights on individual phrase pairs rather than sentences.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For example, if Barry is not in commonWords and is found in the list of person first names, then the feature PersonFirstName will be set to 1.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"In order to train this error-correction model, he divided his training corpus into 5 portions of 20% each.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The effect of UNIQ is very small on both data sets.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We then extract a set of possible tags tx(y) by eliminating labels whose probability is below a threshold value \u03c4: We describe how we choose \u03c4 in \u00a76.4.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The use of ILP in learning the desired grammar significantly increases the computational complexity of this method.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Contextual role knowledge provides evidence as to whether a candidate is a plausible antecedent for an anaphor.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The availability of these resources guided our selection of foreign languages.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"SEGTok(noH) is the segmentation accuracy ignoring mistakes involving the implicit definite article h.11 To evaluate our performance on the tagging task we report CPOS and FPOS corresponding to coarse- and fine-grained PoS tagging results (F1) measure.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Time for Moses itself to load, including loading the language model and phrase table, is included.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Our approach uses word-to-word dependencies between source and target words.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We have not yet tried this.","label":0},{"source_text":"There is no global pruning.","target_text":"In order to handle the necessary word reordering as an optimization problem within our dynamic programming approach, we describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming (Held, Karp, 1962).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Overview of the method 2.2 Step by Step Algorithm.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"BABAR uses a DempsterShafer decision model (Stefik, 1995) to combine the evidence provided by the knowledge sources.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kind of cue.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"One class comprises words derived by productive morphologi\u00c2\u00ad cal processes, such as plural noun formation using the suffix ir, menD.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The accuracies for link were 73% and 86% on two evaluated domains.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This leads to a linear combination of domain-specific probabilities, with weights in [0, 1], normalized to sum to 1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The breakdown of the different types of words found by ST in the test corpus is given in Table 3.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(e.g., N.Y. would contribute this feature, IBM would not). nonalpha=x Appears if the spelling contains any characters other than upper or lower case letters.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The use of the Good-Turing equation presumes suitable estimates of the unknown expectations it requires.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Caching for IRSTLM is smaller at 0.09 GB resident memory, though it supports only a single thread.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The simplest approach involves scoring the various analyses by costs based on word frequency, and picking the lowest cost path; variants of this approach have been described in Chang, Chen, and Chen (1991) and Chang and Chen (1993).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"collected too.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"These two principles guide experimentation in this framework, and together with the evaluation measures help us decide which specific type of substructure to combine.","label":0},{"source_text":"There is no global pruning.","target_text":"(f1; ;mg n fl1; l2g ; l) 4 (f1; ;m \u00f4\u0080\u0080\u0080 1g n fl1; l2; l3g ; l0) !","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"This can be repeated several times to collect a list of author \/ book title pairs and expressions.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In this example there are four \"input characters,\" A, B, C and D, and these map respectively to four \"pronunciations\" a, b, c and d. Furthermore, there are four \"words\" represented in the dictionary.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Otherwise, the scope of the search problem shrinks recursively: if A[pivot] < k then this becomes the new lower bound: l +\u2014 pivot; if A[pivot] > k then u +\u2014 pivot.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We use two common techniques, hash tables and sorted arrays, describing each before the model that uses the technique.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For example, in Figure 3, we can see that the phrases in the \u00e2\u0080\u009cbuy\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire\u00e2\u0080\u009d and \u00e2\u0080\u009cpurchase\u00e2\u0080\u009d sets are mostly paraphrases.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Quantization can be improved by jointly encoding probability and backoff.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(b) F.i'JJI!","label":0},{"source_text":"Here we present two algorithms.","target_text":"Using the virtual distribution Di (i) and pseudo-labels&quot;y.,\u201e values for Wo, W\u00b1 and W_ can be calculated for each possible weak hypothesis (i.e., for each feature x E Xi); the weak hypothesis with minimal value for Wo + 2\/WW _ can be chosen as before; and the weight for this weak hypothesis at = ln ww+411:) can be calculated.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"RandLM and SRILM also remove context that will not extend, but SRILM performs a second lookup in its trie whereas our approach has minimal additional cost.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"An important subproblem of language model storage is therefore sparse mapping: storing values for sparse keys using little memory then retrieving values given keys using little time.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"(2009).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Therefore, for n-gram wn1 , all leftward extensions wn0 are an adjacent block in the n + 1-gram array.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"In contrast, NNP (proper nouns) form a large portion of vocabulary.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Two sets of examples from Gan are given in (1) and (2) (:::::: Gan's Appendix B, exx.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Given around 90,000 unlabeled examples, the methods described in this paper classify names with over 91% accuracy.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Mi(c) is a binary function returning t when parser i (from among the k parsers) suggests constituent c should be in the parse.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Our primary goal is to exploit the resources that are most appropriate for the task at hand, and our secondary goal is to allow for comparison of our models\u2019 performance against previously reported results.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This result suggests that all of contextual role KSs can provide useful information for resolving anaphora.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The transition from f, to a final state transduces c to the grammatical tag PL with cost cost(unseen(f,)): cost(i\u00c2\u00a5JJ1l.ir,) == cost(i\u00c2\u00a5JJ1l.)","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Our results show that BABAR achieves good performance in both domains, and that the contextual role knowledge improves performance, especially on pronouns.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"While building a machine translation system is a serious undertaking, in future we hope to attract more newcomers to the field by keeping the barrier of entry as low as possible.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"We then evaluate the approach in two steps.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, as we have noted, nothing inherent in the approach precludes incorporating higher-order constraints, provided they can be effectively modeled within a finite-state framework.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"For example, the two NEs \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d and \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d have the following contexts.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We report results for the best and median hyperparameter settings obtained in this way.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Table 2 compares the performance of our system on the setup of Cohen and Smith (2007) to the best results reported by them for the same tasks.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"This number must be less than or equal to n \u00f4\u0080\u0080\u0080 1.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"They are set to fixed constants.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Previous work deals with this problem by correcting inconsistencies between the named entity classes assigned to different occurrences of the same entity (Borthwick, 1999; Mikheev et al., 1998).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In the next section we present an alternative approach that builds two classifiers while attempting to satisfy the above constraints as much as possible.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Note that on some examples (around 2% of the test set) CoBoost abstained altogether; in these cases we labeled the test example with the baseline, organization, label.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Thus, we feel fairly confident that for the examples we have considered from Gan's study a solution can be incorporated, or at least approximated, within a finite-state framework.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The family name set is restricted: there are a few hundred single-hanzi family names, and about ten double-hanzi ones.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"It was filtered to retain the top 30 translations for each source phrase using the TM part of the current log-linear model.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Note that the sets of possible classifiers for a given noun can easily be encoded on that noun by grammatical features, which can be referred to by finite-state grammatical rules.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This supports our main thesis that decisions taken by single, improved, grammar are beneficial for both tasks.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Since our objective is to compare distributions of bracketing discrepancies, we do not use heuristics to prune the set of nuclei.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We can use the preceding linguistic and annotation insights to build a manually annotated Arabic grammar in the manner of Klein and Manning (2003).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The government has to make a decision, and do it quickly.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Several extensions of AdaBoost for multiclass problems have been suggested (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Bikel 75 training trees 5000 10000 15000 Figure 3: Dev set learning curves for sentence lengths \u00e2\u0089\u00a4 70.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Upon identifying an anaphoric expression (currently restricted to: pronouns, prepositional adverbs, definite noun phrases), the an- notator first marks the antecedent expression (currently restricted to: various kinds of noun phrases, prepositional phrases, verb phrases, sentences) and then establishes the link between the two.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Word Re-ordering and DP-based Search in Statistical Machine Translation","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For each domain, we created a blind test set by manually annotating 40 doc uments with anaphoric chains, which represent sets of m3 (S) = ) X \u00e2\u0088\u00a9Y =S 1 \u00e2\u0088\u0092 ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) (1) noun phrases that are coreferent (as done for MUC6 (MUC6 Proceedings, 1995)).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"To define a similarity function between the English and the foreign vertices, we rely on high-confidence word alignments.","label":0},{"source_text":"There is no global pruning.","target_text":"An inverted alignment is defined as follows: inverted alignment: i ! j = bi: Target positions i are mapped to source positions bi.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"However, the learning curves in Figure 3 show that the Berkeley parser does not exceed our manual grammar by as wide a margin as has been shown for other languages (Petrov, 2009).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We have shown the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The result of this is shown in Figure 7.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"One hopes that such a corpus will be forth\u00c2\u00ad coming.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"For all languages we do not make use of a tagging dictionary.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Reference resolution involves finding words that co-refer to the same entity.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Thus, the arc from je to jedna will be labeled 5b\u2193 (to indicate that there is a syntactic head below it).","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Even this may be nondeterministic, in case the graph contains several non-projective arcs whose lifts interact, but we use the following algorithm to construct a minimal projective transformation D0 = (W, A0) of a (nonprojective) dependency graph D = (W, A): The function SMALLEST-NONP-ARC returns the non-projective arc with the shortest distance from head to dependent (breaking ties from left to right).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each candidate antecedent, BABAR identifies the caseframe that would extract the candidate, pairs it with the anaphor\u00e2\u0080\u0099s caseframe, and consults the CF Network to see if this pair of caseframes has co-occurred in previous resolutions.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"4 Traditional Arabic linguistic theory treats both of these types as subcategories of noun \u00ef\u00bf\u00bd '.i . Figure 1: The Stanford parser (Klein and Manning, 2002) is unable to recover the verbal reading of the unvocalized surface form 0 an (Table 1).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"As we shall argue, the semantic class affiliation of a hanzi constitutes useful information in predicting its properties.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"English parsing evaluations usually report results on sentences up to length 40.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Instead, we resort to an iterative update based method.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We use w erations of sampling (see Figure 2 for a depiction).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"95 B a s e li n e ( S e lf t a g ) 70 a l l B i k e l ( v 1 . 2 ) B a s e l i n e ( P r e t a g ) 7 0 a l l G o l d P O S 70 0.7 70 0.801 278 0.7 52 0.794 278 0.7 71 0.804 295 0.7 52 0.796 295 0.7 75 0.808 309 77.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Unlike the WSJ corpus which has a high frequency of rules like VP \u00e2\u0086\u0092VB PP, Arabic verb phrases usually have lexi calized intervening nodes (e.g., NP subjects and direct objects).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Thus we are interested not in extraction, but actual generation from representations that may be developed to different degrees of granularity.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This paper does not necessarily reflect the position of the U.S. Government.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For example, if is found in the list of person first names, the feature PersonFirstName is set to 1.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each resolution in the training data, BABAR also associates the co-referring expression of an NP with the NP\u00e2\u0080\u0099s caseframe.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The first modification \u2014 cautiousness \u2014 is a relatively minor change.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The orthographic normalization strategy we use is simple.10 In addition to removing all diacritics, we strip instances of taTweel J=J4.i, collapse variants of alif to bare alif,11 and map Ara bic punctuation characters to their Latin equivalents.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We empirically control for the effect of our heuristics to make sure our pruning does not undermine the objectives of our joint task.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The maximum precision oracle is an upper bound on the possible gain we can achieve by parse hybridization.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"i=1 (f,v)\u00e2\u0088\u0088Wi","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We present two algorithms.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"As such, global information from the whole context of a document is important to more accurately recognize named entities.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Topicalization of NP subjects in SVO configurations causes confusion with VO (pro-drop).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Statistics are printed before Moses exits and after parts of the decoder have been destroyed.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Features and context were initially introduced into the models, but they refused to offer any gains in performance.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For IE, the system must be able to distinguish between semantically similar noun phrases that play different roles in an event.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Compared to the EMEA\/EP setting, the two domains in the NIST setting are less homogeneous and more similar to each other; there is also considerably more IN text available.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"However, lazy mapping is generally slow because queries against uncached pages must wait for the disk.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Each learner is free to pick the labels for these instances.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"However, this result is consistent with the results of ex\u00c2\u00ad periments discussed in Wu and Fung (1994).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"ICOC and CSPP contributed the greatest im provements.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"3 60.7 50.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"As such, global information from the whole context of a document is important to more accurately recognize named entities.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A secondary reference resolution classifier has information on the class assigned by the primary classifier.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"10) and trained both EM and L-BFGS for 1000 iterations.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We first make use of our morphological analyzer to find all segmentation possibilities by chopping off all prefix sequence possibilities (including the empty prefix) and construct a lattice off of them.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Zone: MUC data contains SGML tags, and a document is divided into zones (e.g., headlines and text zones).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"(2003), which gives 96.8% accuracy on the test set.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Experiments are presented in section 4.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"For our experiments we also report the mean of precision and recall, which we denote by (P + R)I2 and F-measure.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Each token may admit multiple analyses, each of which a sequence of one or more lexemes (we use li to denote a lexeme) belonging a presupposed Hebrew lexicon LEX.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The estimation of the probabilities in the model is carried out as shown in Equation 4.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Our TRIE implementation is designed to improve upon IRSTLM using a reverse trie with improved search, bit level packing, and stateful queries.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Two issues distinguish the various proposals.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In Section 5, we try to compare results of MENE, IdentiFinder, and MENERGI.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"There any many techniques for improving language model speed and reducing memory consumption.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Indeed there are several open issues.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"When the same token is to be interpreted as a single lexeme fmnh, it may function as a single adjective \u201cfat\u201d.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We also report word type level accuracy, the fraction of word types assigned their majority tag (where the mapping between model state and tag is determined by greedy one-to-one mapping discussed above).5 For each language, we aggregate results in the following way: First, for each hyperparameter setting, evaluate three variants: The first model (1TW) only 4 Typically, the performance stabilizes after only 10 itera-.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"However, it is unlikely that other occurrences of News Broadcasting Corp. in the same document also co-occur with Even.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We first note that the accuracy results of our system are overall higher on their setup, on all measures, indicating that theirs may be an easier dataset.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"BerkeleyLM revision 152 (Pauls and Klein, 2011) implements tries based on hash tables and sorted arrays in Java with lossy quantization.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The use of weighted transducers in particular has the attractive property that the model, as it stands, can be straightforwardly interfaced to other modules of a larger speech or natural language system: presumably one does not want to segment Chinese text for its own sake but instead with a larger purpose in mind.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Recently, statistical NERs have achieved results that are comparable to hand-coded systems.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"To evaluate proper-name identification, we randomly se\u00c2\u00ad lected 186 sentences containing 12,000 hanzi from our test corpus and segmented the text automatically, tagging personal names; note that for names, there is always a sin\u00c2\u00ad gle unambiguous answer, unlike the more general question of which segmentation is correct.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Here we propose a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"These operations, as we see below, are restricted to be size preserving (as in the case of concatenation in CFG) which implies that they will be linear and non-erasing.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The best analysis of the corpus is taken to be the true analysis, the frequencies are re-estimated, and the algorithm is repeated until it converges.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The 8 similarity-to-IN features are based on word frequencies and scores from various models trained on the IN corpus: To avoid numerical problems, each feature was normalized by subtracting its mean and dividing by its standard deviation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, management succession systems must distinguish between a person who is fired and a person who is hired.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We realize the importance of paraphrase; however, the major obstacle is the construction of paraphrase knowledge.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"8 1 2.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"A named entity recognizer (NER) is useful in many NLP applications such as information extraction, question answering, etc. On its own, a NER can also provide users who are looking for person or organization names with quick information.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Second, we treat the projected labels as features in an unsupervised model (\u00a75), rather than using them directly for supervised training.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"While possible to utilize the feature-based log-linear approach described in Berg-Kirkpatrick et al.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"na me =>2 ha nzi fa mi ly 1 ha nzi gi ve n 6.1 ha nzi fa mi ly => ha nz ii 7.2 ha nzi fa mi ly => ha nzi i ha nz ij 8.1 ha nzi gi ve n => ha nz ii 9.2 ha nzi giv en => ha nzi i ha nz ij The difficulty is that given names can consist, in principle, of any hanzi or pair of hanzi, so the possible given names are limited only by the total number of hanzi, though some hanzi are certainly far more likely than others.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Ex: The brigade, which attacked ...","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Motivated by these questions, we significantly raise baselines for three existing parsing models through better grammar engineering.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We consider two variants of Berg-Kirkpatrick et al.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Nodes in the trie are based on arrays sorted by vocabulary identifier.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A Brief Introduction to the Chinese Writing System Most readers will undoubtedly be at least somewhat familiar with the nature of the Chinese writing system, but there are enough common misunderstandings that it is as well to spend a few paragraphs on properties of the Chinese script that will be relevant to topics discussed in this paper.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(3) shows learning curves for CoBoost.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In our experiment, we set the threshold of the TF\/ITF score empirically using a small development corpus; a finer adjustment of the threshold could reduce the number of such keywords.","label":0},{"source_text":"There is no global pruning.","target_text":"The experimental tests are carried out on the Verbmobil task (GermanEnglish, 8000-word vocabulary), which is a limited-domain spoken-language task.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The cost estimate, cost(i\u00c2\u00a5JJ1l.fn is computed in the obvious way by summing the negative log probabilities of i\u00c2\u00a5JJ1l.","label":0},{"source_text":"Here we present two algorithms.","target_text":"This section describes an algorithm based on boosting algorithms, which were previously developed for supervised machine learning problems.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"(Bar-Haim et al., 2007; Habash and Rambow, 2005)) and probabilities are assigned to different analyses in accordance with the likelihood of their tags (e.g., \u201cfmnh is 30% likely to be tagged NN and 70% likely to be tagged REL+VB\u201d).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"ATB CTB6 Negra WSJ Trees 23449 28278 20602 43948 Word Typess 40972 45245 51272 46348 Tokens 738654 782541 355096 1046829 Tags 32 34 499 45 Phrasal Cats 22 26 325 27 Test OOV 16.8% 22.2% 30.5% 13.2% Per Sentence Table 4: Gross statistics for several different treebanks.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In this work we extended the AdaBoost.MH (Schapire and Singer 98) algorithm to the cotraining case.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The model is often further restricted so that each source word is assigned to exactly one target word (Brown et al., 1993; Ney et al., 2000).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"9 50.2 +P RI OR be st me dia n 47.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Indeed, as we shall show in Section 5, even human judges differ when presented with the task of segmenting a text into words, so a definition of the criteria used to determine that a given segmentation is correct is crucial before one can interpret such measures.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"However, there will remain a large number of words that are not readily adduced to any produc\u00c2\u00ad tive pattern and that would simply have to be added to the dictionary.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"J:j:l :zhongl :0.0 ;m,Jlong4 :0.0 (mHHaryg9tltHBI) \u00c2\u00a3: _ADV: 5.98 \u00c2\u00a5 :hua2:o.o E :_NC: 4.41 :mln2:o.o mm : guo2 : 0.0 (RopubllcofChlna) .....,.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Pr(eI 1) is the language model of the target language, whereas Pr(fJ 1 jeI1) is the transla tion model.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"A detailed description of the search procedure used is given in this patent.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Furthermore, we know one of the original parses will be the hypothesized parse, so the direct method of determining which one is best is to compute the probability of each of the candidate parses using the probabilistic model we developed in Section 2.1.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Much could be done to further reduce memory consumption.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If somewhere else in the document we see \u00e2\u0080\u009crestrictions put in place by President Bush\u00e2\u0080\u009d, then we can be surer that Bush is a name.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This is consistent with the nature of these two settings: log-linear combination, which effectively takes the intersection of IN and OUT, does relatively better on NIST, where the domains are broader and closer together.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, there is again local grammatical information that should favor the split in the case of (1a): both .ma3 'horse' and .ma3 lu4 are nouns, but only .ma3 is consistent with the classifier pil, the classifier for horses.21 By a similar argument, the preference for not splitting , lm could be strengthened in (lb) by the observation that the classifier 1'1* tiao2 is consistent with long or winding objects like , lm ma3lu4 'road' but not with,ma3 'horse.'","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The parser builds dependency graphs by traversing the input from left to right, using a stack to store tokens that are not yet complete with respect to their dependents.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The second main result is that the pseudo-projective approach to parsing (using special arc labels to guide an inverse transformation) gives a further improvement of about one percentage point on attachment score.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"A token that is allCaps will also be initCaps.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Both authors are members of the Center for Language and Speech Processing at Johns Hopkins University.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"1 2 3.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For eight judges, ranging k between 1 and 8 corresponded to a precision score range of 90% to 30%, meaning that there were relatively few words (30% of those found by the automatic segmenter) on which all judges agreed, whereas most of the words found by the segmenter were such that one human judge agreed.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We use a patched version of BitPar allowing for direct input of probabilities instead of counts.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We plan to explore more powerful techniques for exploiting the diversity of parsing methods.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This combination generalizes (2) and (3): we use either at = a to obtain a fixed-weight linear combination, or at = cI(t)\/(cI(t) + 0) to obtain a MAP combination.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Each ht is a function that predicts a label (+1 or \u20141) on examples containing a particular feature xt, while abstaining on other examples: The prediction of the strong hypothesis can then be written as We now briefly describe how to choose ht and at at each iteration.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Our TRIE implementation is designed to improve upon IRSTLM using a reverse trie with improved search, bit level packing, and stateful queries.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"3 60.7 50.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The composition operations in the case of CFG's are parameterized by the productions.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"If evidence indicates that hypotheses C and D are less likely than hypotheses A and B, then probabilities are redistributed to reflect the fact that {A, B} is more likely to contain the answer than {C, D}.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"level.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The corpus was wordaligned using both HMM and IBM2 models, and the phrase table was the union of phrases extracted from these separate alignments, with a length limit of 7.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Clearly, for judges h and h taking h as standard and computing the precision and recall for Jz yields the same results as taking h as the standard, and computing for h, 14 All evaluation materials, with the exception of those used for evaluating personal names were drawn.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Despite these limitations, a purely finite-state approach to Chinese word segmentation enjoys a number of strong advantages.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The monotone search performs worst in terms of both error rates mWER and SSER.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Similar behavior is observed when adding features.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"With regard to purely morphological phenomena, certain processes are not han\u00c2\u00ad dled elegantly within the current framework Any process involving reduplication, for instance, does not lend itself to modeling by finite-state techniques, since there is no way that finite-state networks can directly implement the copying operations required.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Altun et al. (2005) proposed a technique that uses graph based similarity between labeled and unlabeled parts of structured data in a discriminative framework for semi-supervised learning.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We present a coreference resolver called BABAR that uses contextual role knowledge to evaluate possible antecedents for an anaphor.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Multiple features can be used for the same token.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"M works in space S(n) if for every string that M accepts no configuration exceeds space S(n).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The P (T |\u00cf\u0088) distribution, in English for instance, should have very low mass for the DT (determiner) tag, since determiners are a very small portion of the vocabulary.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"3.4 Salience-based text generation.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The 1-bit sign is almost always negative and the 8-bit exponent is not fully used on the range of values, so in practice this corresponds to quantization ranging from 17 to 20 total bits.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The original tag set for the CoNLL-X Dutch data set consists of compounded tags that are used to tag multi-word units (MWUs) resulting in a tag set of over 300 tags.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"was done by the participants.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This has solutions: where pI(s|t) is derived from the IN corpus using relative-frequency estimates, and po(s|t) is an instance-weighted model derived from the OUT corpus.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Several papers report the use of part-of-speech information to rank segmentations (Lin, Chiang, and Su 1993; Peng and Chang 1993; Chang and Chen 1993); typically, the probability of a segmentation is multiplied by the probability of the tagging(s) for that segmentation to yield an estimate of the total probability for the analysis.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"In practice, this sparsity constraint is difficult to incorporate in a traditional POS induction system (Me\u00c2\u00b4rialdo, 1994; Johnson, 2007; Gao and Johnson, 2008; Grac\u00c2\u00b8a et al., 2009; Berg-Kirkpatrick et al., 2010).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In the small window on the left, search queries can be entered, here one for an NP that has been annotated on the co-reference layer as bridging.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"36 79.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In the maximum entropy framework, there is no such constraint.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Case and Zone: If the token starts with a capital letter (initCaps), then an additional feature (init- Caps, zone) is set to 1.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Using structural information As was explained in the results section, we extracted examples like \u00e2\u0080\u009cSmith estimates Lotus\u00e2\u0080\u009d, from a sentence like \u00e2\u0080\u009cMr.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Because we are interested in applying our techniques to languages for which no labeled resources are available, we paid particular attention to minimize the number of free parameters and used the same hyperparameters for all language pairs.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"These are shown, with their associated costs, as follows: ABj nc 4.0 AB C\/jj 6.0 CD \/vb 5.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We claim that no particular morphological segmentation is a-priory more likely for surface forms before exploring the compositional nature of syntactic structures, including manifestations of various long-distance dependencies.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"(2009) on Portuguese (Grac\u00c2\u00b8a et al.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Training examples are generated automatically by identifying noun phrases that can be easily resolved with their antecedents using lexical and syntactic heuristics.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"36.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Evaluation results within sets Table 1 shows the evaluation result based on the number of phrases in a set.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We can define derivation trees inductively on the length of the derivation of a tree 1.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"encodes the one tag per word constraint and is uni form over type-level tag assignments.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"A summary of the corpus used in the experiments is given in Table 3.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"There is a fairly large body of work on SMT adaptation.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"A procedural definition to restrict1In the approach described in (Berger et al., 1996), a mor phological analysis is carried out and word morphemes rather than full-form words are used during the search.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"\u00e2\u0080\u0094 similar results have been observed across multiple languages.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Finally, this effort is part of a much larger program that we are undertaking to develop stochastic finite-state methods for text analysis with applications to TIS and other areas; in the final section of this paper we will briefly discuss this larger program so as to situate the work discussed here in a broader context.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Trying to integrate constituent ordering and choice of referring expressions, (Chiarcos 2003) developed a numerical model of salience propagation that captures various factors of author\u00e2\u0080\u0099s intentions and of information structure for ordering sentences as well as smaller constituents, and picking appropriate referring expressions.10 Chiarcos used the PCC annotations of co-reference and information structure to compute his numerical models for salience projection across the generated texts.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"\u00ce\u00b2 is the shared hyperparameter for the tag assignment prior and word feature multinomials.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In these experiments, the input lacks segmentation markers, hence the slightly different dev set baseline than in Table 6.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We used it to score all phrase pairs in the OUT table, in order to provide a feature for the instance-weighting model.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If is unique, then a feature (Unique, Zone) is set to 1, where Zone is the document zone where appears.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For pronouns, however, all of the knowledge sources increased recall, often substantially, and with little if any decrease in precision.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A modified language model probability p\u00c3\u0086(eje0; e00) is defined as follows: p\u00c3\u0086(eje0; e00) =  1:0 if \u00c3\u0086 = 0 p(eje0; e00) if \u00c3\u0086 = 1 : We associate a distribution p(\u00c3\u0086) with the two cases \u00c3\u0086 = 0 and \u00c3\u0086 = 1 and set p(\u00c3\u0086 = 1) = 0:7.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Each pattern represents the role that a noun phrase plays in the surrounding context.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Mandarin exhibits several such processes, including A-not-A question formation, il\u00c2\u00ad lustrated in (3a), and adverbial reduplication, illustrated in (3b): 3.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"rently, some annotations (in particular the connectives and scopes) have already moved beyond the core corpus; the others will grow step by step.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We obtained positive results using a very simple phrase-based system in two different adaptation settings: using English\/French Europarl to improve a performance on a small, specialized medical domain; and using non-news portions of the NIST09 training material to improve performance on the news-related corpora.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"), which precludes a single universal approach to adaptation.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We do not attempt to identify the types of relationships that are found.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"3.4 Salience-based text generation.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"LCFRS's share several properties possessed by the class of mildly context-sensitive formalisms discussed by Joshi (1983\/85).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In this section, we describe how contextual role knowledge is represented and learned.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For RandLM, we used the settings in the documentation: 8 bits per value and false positive probability 1 256.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Figure 2: An ATB sample from the human evaluation.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The dictionary sizes reported in the literature range from 17,000 to 125,000 entries, and it seems reasonable to assume that the coverage of the base dictionary constitutes a major factor in the performance of the various approaches, possibly more important than the particular set of methods used in the segmentation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"handled given appropriate models.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"encodes the one tag per word constraint and is uni form over type-level tag assignments.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We used this data to build an unpruned ARPA file with IRSTLM\u2019s improved-kneser-ney option and the default three pieces.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We report results for the best and median hyperparameter settings obtained in this way.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Assume that the two classifiers are &quot;rote learners&quot;: that is, 1.1 and 12 are defined through look-up tables that list a label for each member of X1 or X2.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"So, we set a threshold that at least two examples are required to build a link.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"42 nator, the N31s can be measured well by counting, and we replace the expectation by the observation.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Case and Zone of and : Similarly, if (or ) is initCaps, a feature (initCaps, zone) (or (initCaps, zone) ) is set to 1, etc. Token Information: This group consists of 10 features based on the string , as listed in Table 1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Other kinds of productive word classes, such as company names, abbreviations (termed fijsuolxie3 in Mandarin), and place names can easily be 20 Note that 7 in E 7 is normally pronounced as leO, but as part of a resultative it is liao3..","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"1 74.5 56.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"3.1 Word ReOrdering with Verbgroup.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"36.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The approach builds from an initial seed set for a category, and is quite similar to the decision list approach described in (Yarowsky 95).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"There has also been work using a bootstrap- ping approach [Brin 98; Agichtein and Gravano 00; Ravichandran and Hovy 02].","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We have already mentioned the closely related work by Matsoukas et al (2009) on discriminative corpus weighting, and Jiang and Zhai (2007) on (nondiscriminative) instance weighting.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Contextual role knowledge provides evidence as to whether a candidate is a plausible antecedent for an anaphor.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"On the other hand, given that all schemes have similar parsing accuracy overall, this means that the Path scheme is the least likely to introduce errors on projective arcs.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In MUC6, the best result is achieved by SRA (Krupka, 1995).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"buy - acquire (5) buy - agree (2) buy - purchase (5) buy - acquisition (7) buy - pay (2)* buy - buyout (3) buy - bid (2) acquire - purchase (2) acquire - acquisition (2) acquire - pay (2)* purchase - acquisition (4) purchase - stake (2)* acquisition - stake (2)* unit - subsidiary (2) unit - parent (5) It is clear that these links form two clusters which are mostly correct.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"31 75.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"(2010)\u00e2\u0080\u0099s richest model: optimized via either EM or LBFGS, as their relative performance depends on the language.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The counts represent portions of the approximately 44000 constituents hypothesized by the parsers in the development set.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The method uses a &quot;soft&quot; measure of the agreement between two classifiers as an objective function; we described an algorithm which directly optimizes this function.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"That is, given a choice between segmenting a sequence abc into abc and ab, c, the former will always be picked so long as its cost does not exceed the summed costs of ab and c: while; it is possible for abc to be so costly as to preclude the larger grouping, this will certainly not usually be the case.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Restrictions: Quasi-monotone Search The above search space is still too large to allow the translation of a medium length input sentence.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The likelihood of the observed data under the model is where P(yi, xi) is defined as in (9).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The following algorithm was then used to induce new rules: Let Count' (x) be the number of times feature x is seen with some known label in the training data.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Segmentation errors cascade into the parsing phase, placing an artificial limit on parsing performance.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"This stage of label propagation results in a tag distribution ri over labels y, which encodes the proportion of times the middle word of ui E Vf aligns to English words vy tagged with label y: The second stage consists of running traditional label propagation to propagate labels from these peripheral vertices Vf\ufffd to all foreign language vertices in the graph, optimizing the following objective: 5 POS Induction After running label propagation (LP), we compute tag probabilities for foreign word types x by marginalizing the POS tag distributions of foreign trigrams ui = x\u2212 x x+ over the left and right context words: where the qi (i = 1, ... , |Vf|) are the label distributions over the foreign language vertices and \u00b5 and \u03bd are hyperparameters that we discuss in \u00a76.4.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Further, the special hash 0 suffices to flag empty buckets.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"This work has been supported as part of the Verbmobil project (contract number 01 IV 601 A) by the German Federal Ministry of Education, Science, Research and Technology and as part of the Eutrans project (ESPRIT project number 30268) by the European Community.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Our results suggest that it is possible to learn accurate POS taggers for languages which do not have any annotated data, but have translations into a resource-rich language.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The development of automatic scoring methods is an open field of research.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In a grammar which generates independent paths the derivations of sibling constituents can not share an unbounded amount of information.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Note also that the costs currently used in the system are actually string costs, rather than word costs.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Papers that use this method or minor variants thereof include Liang (1986), Li et al.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"As we will see from Table 3, not much improvement is derived from this feature.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Assigning rhetorical relations thus poses questions that can often be answered only subjectively.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We introduce several new ideas.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The model is often further restricted so that each source word is assigned to exactly one target word (Brown et al., 1993; Ney et al., 2000).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The similar explanation applies to the link to the \u00e2\u0080\u009cstake\u00e2\u0080\u009d set.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"However, when the linear LM is combined with a linear TM (lm+lin tm) or MAP TM (lm+map TM), the results are much better than a log-linear combination for the EMEA setting, and on a par for NIST.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"While sorted arrays could be used to implement the same data structure as PROBING, effectively making m = 1, we abandoned this implementation because it is slower and larger than a trie implementation.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The semantic caseframe expectations are used in two ways.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"IRSTLM (Federico et al., 2008) is an open-source toolkit for building and querying language models.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In Section 4, we present the performance measures used and give translation results on the Verbmobil task.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We briefly review it here for completeness.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Using the virtual distribution Di (i) and pseudo-labels&quot;y.,\u201e values for Wo, W\u00b1 and W_ can be calculated for each possible weak hypothesis (i.e., for each feature x E Xi); the weak hypothesis with minimal value for Wo + 2\/WW _ can be chosen as before; and the weight for this weak hypothesis at = ln ww+411:) can be calculated.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"However, they list two sets, one consisting of 28 fragments and the other of 22 fragments, in which they had 0% recall and precision.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"However, lexically similar NPs usually refer to the same entity in two cases: proper names and existential noun phrases.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Of course, since the number of attested (phonemic) Mandarin syllables (roughly 1400, including tonal distinctions) is far smaller than the number of morphemes, it follows that a given syllable could in principle be written with any of several different hanzi, depending upon which morpheme is intended: the syllable zhongl could be lfl 'middle,''clock,''end,' or ,'loyal.'","label":0},{"source_text":"Here we present two algorithms.","target_text":"Our first algorithm is similar to Yarowsky's, but with some important modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"There is no global pruning.","target_text":"3) A tight coupling with the speech recognizer output.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"A sentence was withheld from section 22 because its extreme length was troublesome for a couple of the parsers.'","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Any opinions, findings, conclusions, or recommendations expressed in this paper are those of the authors, and do not necessarily reflect the views of the funding organizations.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"raphy: A ren2 'person' is a fairly uncontroversial case of a monographemic word, and rplil zhong1guo2 (middle country) 'China' a fairly uncontroversial case of a di\u00c2\u00ad graphernic word.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"We developed a first version of annotation guidelines for co-reference in PCC (Gross 2003), which served as basis for annotating the core corpus but have not been empirically evaluated for inter-annotator agreement yet.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The results are given in Table 4.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Feature selection is implemented using a feature cutoff: features seen less than a small count during training will not be used.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We tabulate this increase in Table 3.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The 2(Yarowsky 95) describes the use of more sophisticated smoothing methods.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The task is to learn a function from an input string (proper name) to its type, which we will assume to be one of the categories Person, Organization, or Location.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The probability of the classes assigned to the words in a sentence in a document is defined as follows: where is determined by the maximum entropy classifier.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Queries detect the invalid probability, using the node only if it leads to a longer match.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"5 64.7 42.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The cost is computed as follows, where N is the corpus size and f is the frequency: (1) Besides actual words from the base dictionary, the lexicon contains all hanzi in the Big 5 Chinese code\/ with their pronunciation(s), plus entries for other characters that can be found in Chinese text, such as Roman letters, numerals, and special symbols.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Cur In order to ground such approaches in linguistic observation and description, a multi-level anno 10 For an exposition of the idea as applied to the task of text planning, see (Chiarcos, Stede 2004).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Instead, we condition on the type-level tag assignments T . Specifically, let St = {i|Ti = t} denote the indices of theword types which have been assigned tag t accord ing to the tag assignments T . Then \u00ce\u00b8t is drawn from DIRICHLET(\u00ce\u00b1, St), a symmetric Dirichlet which only places mass on word types indicated by St. This ensures that each word will only be assigned a single tag at inference time (see Section 4).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We incorporate instance weighting into a mixture-model framework, and find that it yields consistent improvements over a wide range of baselines.","label":1},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The estimation of the probabilities in the model is carried out as shown in Equation 4.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"buy - acquire (5) buy - agree (2) buy - purchase (5) buy - acquisition (7) buy - pay (2)* buy - buyout (3) buy - bid (2) acquire - purchase (2) acquire - acquisition (2) acquire - pay (2)* purchase - acquisition (4) purchase - stake (2)* acquisition - stake (2)* unit - subsidiary (2) unit - parent (5) It is clear that these links form two clusters which are mostly correct.","label":0},{"source_text":"There is no global pruning.","target_text":"1 is given in Fig.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For every sequence of initial capitalized words, its longest substring that occurs in the same document as a sequence of initCaps is identified.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Each token may admit multiple analyses, each of which a sequence of one or more lexemes (we use li to denote a lexeme) belonging a presupposed Hebrew lexicon LEX.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This highly effective approach is not directly applicable to the multinomial models used for core SMT components, which have no natural method for combining split features, so we rely on an instance-weighting approach (Jiang and Zhai, 2007) to downweight domain-specific examples in OUT.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This is because in many cases, the use of hyphens can be considered to be optional (e.g., third-quarter or third quarter).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Each decision determines the inclusion or exclusion of a candidate constituent.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"For seven out of eight languages a threshold of 0.2 gave the best results for our final model, which indicates that for languages without any validation set, r = 0.2 can be used.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If is seen infrequently during training (less than a small count), then will not be selected as a feature and all features in this group are set to 0.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Thus an explicit assumption about the redundancy of the features \u2014 that either the spelling or context alone should be sufficient to build a classifier \u2014 has been built into the algorithm.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Here a set is represented by the keyword and the number in parentheses indicates the number of shared NE pair instances.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"In this case we are interested in finding' the maximum probability parse, ri, and Mi is the set of relevant (binary) parsing decisions made by parser i. ri is a parse selected from among the outputs of the individual parsers.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The key to the methods we describe is redundancy in the unlabeled data.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"But the city name Sharm Al- Sheikh is also iDafa, hence the possibility for the incorrect annotation in (b).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Specifically, the lexicon is generated as: P (T , W |\u00cf\u0088) =P (T )P (W |T ) Word Type Features (FEATS): Past unsupervised POS work have derived benefits from features on word types, such as suffix and capitalization features (Hasan and Ng, 2009; Berg-Kirkpatrick et al.,2010).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Then, it can be verified that We can now derive the CoBoost algorithm as a means of minimizing Zco.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If is unique, then a feature (Unique, Zone) is set to 1, where Zone is the document zone where appears.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Email: rlls@bell-labs.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For displaying and querying the annoated text, we make use of the Annis Linguistic Database developed in our group for a large research effort (\u00e2\u0080\u0098Sonderforschungsbereich\u00e2\u0080\u0099) revolving around 9 2.7 Information structure.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Particular instances of relations are associated with goodness scores.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This suggests a strategy: run interpolation search until the range narrows to 4096 or fewer entries, then switch to binary search.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Local features are features that are based on neighboring tokens, as well as the token itself.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"If so, the CF Network reports that the anaphor and candidate may be coreferent.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Sheffield Guthrie and Hepple (2010) explore several randomized compression techniques, but did not release code.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"(2009) also report results on English, but on the reduced 17 tag set, which is not comparable to ours).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"21 In Chinese, numerals and demonstratives cannot modify nouns directly, and must be accompanied by.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Set the decision list to include all rules whose (smoothed) strength is above some threshold Pmin.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"of Arabic.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The second row represents the performance of the median hyperparameter setting.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"For Experiment 1 it is meaningless as a baseline, since it would result in 0% accuracy. mation on path labels but drop the information about the syntactic head of the lifted arc, using the label d\u2191 instead of d\u2191h (AuxP\u2191 instead of AuxP\u2191Sb).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Future work should also extend the approach to build a complete named entity extractor - a method that pulls proper names from text and then classifies them.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"To establish a soft correspondence between the two languages, we use a second similarity function, which leverages standard unsupervised word alignment statistics (\u00a73.3).3 Since we have no labeled foreign data, our goal is to project syntactic information from the English side to the foreign side.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The corpora for both settings are summarized in table 1.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We will refer to our system as MENERGI (Maximum Entropy Named Entity Recognizer using Global Information).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Our clue is the NE instance pairs.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Benchmarks use the package\u2019s binary format; our code is also the fastest at building a binary file.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The system will then look for sequences of initial capitalized words that match the acronyms found in the whole document.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For each case- frame, BABAR collects the head nouns of noun phrases that were extracted by the caseframe in the training corpus.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Evaluating parsing results in our joint framework, as argued by Tsarfaty (2006), is not trivial under the joint disambiguation task, as the hypothesized yield need not coincide with the correct one.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"10.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"These tag distributions are used to initialize the label distributions over the English vertices in the graph.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"For example, the formalisms in the hierarchy described above generate semilinear languages although their path sets become increasingly more complex as one moves up the hierarchy.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"4.2 Global Features.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This is to allow for fair comparison between the statistical method and GR, which is also purely dictionary-based.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"On several languages, we report performance exceeding that of more complex state-of-the art systems.1","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We can make several observations on the cause of errors.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Evaluation of Morphological Analysis.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"One hybridization strategy is to let the parsers vote on constituents' membership in the hypothesized set.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The numbers falling into the location, person, organization categories were 186, 289 and 402 respectively.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"If (wi, r, wj) E A, we say that wi is the head of wj and wj a dependent of wi.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This is an issue that we have not addressed at the current stage of our research.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Several papers report the use of part-of-speech information to rank segmentations (Lin, Chiang, and Su 1993; Peng and Chang 1993; Chang and Chen 1993); typically, the probability of a segmentation is multiplied by the probability of the tagging(s) for that segmentation to yield an estimate of the total probability for the analysis.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"30 75.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Given a PCFG grammar G and a lattice L with nodes n1 ... nk, we construct the weighted grammar GL as follows: for every arc (lexeme) l E L from node ni to node nj, we add to GL the rule [l --+ tni, tni+1, ... , tnj_1] with a probability of 1 (this indicates the lexeme l spans from node ni to node nj).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Since Dane\u00cb\u0087s\u00e2\u0080\u0099 proposals of \u00e2\u0080\u0098thematic development patterns\u00e2\u0080\u0099, a few suggestions have been made as to the existence of a level of discourse structure that would predict the information structure of sentences within texts.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":".","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Although the tag distributions of the foreign words (Eq.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"German English Training: Sentences 58 073 Words 519 523 549 921 Words* 418 979 453 632 Vocabulary Size 7939 4648 Singletons 3454 1699 Test-147: Sentences 147 Words 1 968 2 173 Perplexity { 26:5 Table 4: Multi-reference word error rate (mWER) and subjective sentence error rate (SSER) for three different search procedures.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In general, several gross corpus statistics favor the ATB, so other factors must contribute to parsing underperformance.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For instance, the common \"suffixes,\" -nia (e.g.,.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(8) can now be rewritten5 as which is of the same form as the function Zt used in AdaBoost.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The CoBoost algorithm described above divides the function Zco into two parts: Zco = 40 + 40.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In this paper, we will propose an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"This may seem surprising, given the experiments reported in section 4, but the explanation is probably that the non-projective dependencies that can be recovered at all are of the simple kind that only requires a single lift, where the encoding of path information is often redundant.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"From Thatcher's (1973) work, it is obvious that the complexity of the set of paths from root to frontier of trees in a local set (the tree set of a CFG) is regular'.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The original tag set for the CoNLL-X Dutch data set consists of compounded tags that are used to tag multi-word units (MWUs) resulting in a tag set of over 300 tags.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The hyperparameters \u00ce\u00b1 and \u00ce\u00b2 represent the concentration parameters of the token- and type-level components of the model respectively.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"This approach is compared to another reordering scheme presented in (Berger et al., 1996).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Intuitively, as suggested by the example in the introduction, this is the right granularity to capture domain effects.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Thus our proposed model is a proper model assigning probability mass to all (7r, L) pairs, where 7r is a parse tree and L is the one and only lattice that a sequence of characters (and spaces) W over our alpha-beth gives rise to.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"All annotations are done with specific tools and in XML; each layer has its own DTD.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"A graphical depiction of our model as well as a summary of random variables and parameters can be found in Figure 1.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"There has also been some work on adapting the word alignment model prior to phrase extraction (Civera and Juan, 2007; Wu et al., 2005), and on dynamically choosing a dev set (Xu et al., 2007).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The model was built with open vocabulary, modified Kneser-Ney smoothing, and default pruning settings that remove singletons of order 3 and higher.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The portions of information in the large window can be individually clicked visible or invisible; here we have chosen to see (from top to bottom) \u00e2\u0080\u00a2 the full text, \u00e2\u0080\u00a2 the annotation values for the activated annotation set (co-reference), \u00e2\u0080\u00a2 the actual annotation tiers, and \u00e2\u0080\u00a2 the portion of text currently \u00e2\u0080\u0098in focus\u00e2\u0080\u0099 (which also appears underlined in the full text).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"For the LM, adaptive weights are set as follows: where \u03b1 is a weight vector containing an element \u03b1i for each domain (just IN and OUT in our case), pi are the corresponding domain-specific models, and \u02dcp(w, h) is an empirical distribution from a targetlanguage training corpus\u2014we used the IN dev set for this.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Can we do . QmS: Yes, wonderful.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"All four of the techniques studied result in parsing systems that perform better than any previously reported.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Thus at each iteration the algorithm is forced to pick features for the location, person and organization in turn for the classifier being trained.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"For each set, the phrases with bracketed frequencies are considered not paraphrases in the set.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"To see this, note thai the first two terms in the above equation correspond to the function that AdaBoost attempts to minimize in the standard supervised setting (Equ.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But the city name Sharm Al- Sheikh is also iDafa, hence the possibility for the incorrect annotation in (b).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Both authors are members of the Center for Language and Speech Processing at Johns Hopkins University.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Yet, some hanzi are far more probable in women's names than they are in men's names, and there is a similar list of male-oriented hanzi: mixing hanzi from these two lists is generally less likely than would be predicted by the independence model.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The approach recursively evaluates a quantity Q(C; j), where C is the set of already visited cities and sj is the last visited city.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The semilinearity of Tree Adjoining Languages (TAL's), MCTAL's, and Head Languages (HL's) can be proved using this property, with suitable restrictions on the composition operations.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"All our results are obtained by using only the official training data provided by the MUC conferences.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, management succession systems must distinguish between a person who is fired and a person who is hired.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"If the key distribution\u2019s range is also known (i.e. vocabulary identifiers range from 0 to the number of words), then interpolation search can use this information instead of reading A[0] and A[|A |\u2212 1] to estimate pivots; this optimization alone led to a 24% speed improvement.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"We are given a source string fJ 1 = f1:::fj :::fJ of length J, which is to be translated into a target string eI 1 = e1:::ei:::eI of length I. Among all possible target strings, we will choose the string with the highest probability: ^eI 1 = arg max eI 1 fPr(eI 1jfJ 1 )g = arg max eI 1 fPr(eI 1) Pr(fJ 1 jeI 1)g : (1) The argmax operation denotes the search problem, i.e. the generation of the output sentence in the target language.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"An easy way to achieve this is to put the domain-specific LMs and TMs into the top-level log-linear model and learn optimal weights with MERT (Och, 2003).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"3.2 Inter-annotator Agreement.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Once HMM parameters (\u00ce\u00b8, \u00cf\u0086) are drawn, a token-level tag and word sequence, (t, w), is generated in the standard HMM fashion: a tag sequence t is generated from \u00cf\u0086.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"German English Training: Sentences 58 073 Words 519 523 549 921 Words* 418 979 453 632 Vocabulary Size 7939 4648 Singletons 3454 1699 Test-147: Sentences 147 Words 1 968 2 173 Perplexity { 26:5 Table 4: Multi-reference word error rate (mWER) and subjective sentence error rate (SSER) for three different search procedures.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The approach gains leverage from natural redundancy in the data: for many named-entity instances both the spelling of the name and the context in which it appears are sufficient to determine its type.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For example, suppose the current model assigns a belief value of .60 to {A, B}, meaning that it is 60% sure that the correct hypothesis is either A or B. Then new evidence arrives with a belief value of .70 assigned 5 Initially there are no competing hypotheses because all hypotheses are included in \u00ce\u00b8 by definition.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"This might be because our features are more comprehensive than those used by Borthwick.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The algorithm in Fig.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Figure 3 shows a small fragment of the WFST encoding the dictionary, containing both entries forjust discussed, g:t\u00c2\u00a5 zhonglhua2 min2guo2 (China Republic) 'Republic of China,' and i\u00c2\u00a5inl.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In this way, the method reported on here will necessarily be similar to a greedy method, though of course not identical.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Finally, since non-projective constructions often involve long-distance dependencies, the problem is closely related to the recovery of empty categories and non-local dependencies in constituency-based parsing (Johnson, 2002; Dienes and Dubey, 2003; Jijkoun and de Rijke, 2004; Cahill et al., 2004; Levy and Manning, 2004; Campbell, 2004).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Because we don\u2019t have a separate development set, we used the training set to select among them and found 0.2 to work slightly better than 0.1 and 0.3.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"For unaligned words, we set the tag to the most frequent tag in the corresponding treebank.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"If one of these checks fails then this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We used C = 1.0 as the L2 regularization constant in (Eq.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"MENE (Maximum Entropy Named Entity) (Borth- wick, 1999) was combined with Proteus (a hand- coded system), and came in fourth among all MUC 7 participants.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Morphological Analyzer Ideally, we would use an of-the-shelf morphological analyzer for mapping each input token to its possible analyses.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Upon identifying an anaphoric expression (currently restricted to: pronouns, prepositional adverbs, definite noun phrases), the an- notator first marks the antecedent expression (currently restricted to: various kinds of noun phrases, prepositional phrases, verb phrases, sentences) and then establishes the link between the two.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The quasi-monotone search performs best in terms of both error rates mWER and SSER.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Nu mb er filters candidate if number doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Section 3 discusses the applications that have been completed with PCC, or are under way, or are planned for the future.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"), and thosethat begin with a verb (\u00ef\u00bf\u00bd ub..i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We used it to score all phrase pairs in the OUT table, in order to provide a feature for the instance-weighting model.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Proper-Name Identification.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"In our model, we associate these features at the type-level in the lexicon.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We do not adapt the alignment procedure for generating the phrase table from which the TM distributions are derived.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Such a classification can be seen as a not-always-correct summary of global features.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Given parameter estimates, the label for a test example x is defined as We should note that the model in equation 9 is deficient, in that it assigns greater than zero probability to some feature combinations that are impossible.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"A less canonical representation of segmental morphology is triggered by a morpho-phonological process of omitting the definite article h when occurring after the particles b or l. This process triggers ambiguity as for the definiteness status of Nouns following these particles.We refer to such cases in which the concatenation of elements does not strictly correspond to the original surface form as super-segmental morphology.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"However, it is robust, efficient, and easy to implement.4 To perform the maximization in (7), we used the popular L-BFGS algorithm (Liu and Nocedal, 1989), which requires gradient information.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"A summary of the corpus used in the experiments is given in Table 3.","label":0},{"source_text":"Here we present two algorithms.","target_text":"This &quot;default&quot; feature type has 100% coverage (it is seen on every example) but a low, baseline precision.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Label propagation can then be used to transfer the labels to the peripheral foreign vertices (i.e. the ones adjacent to the English vertices) first, and then among all of the foreign vertices (\u00a74).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"A simple extension will be used to handle this problem.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Therefore in cases where the segmentation is identical between the two systems we assume that tagging is also identical.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We plan to explore more powerful techniques for exploiting the diversity of parsing methods.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The likelihood of the observed data under the model is where P(yi, xi) is defined as in (9).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Based on revision 4041, we modified Moses to print process statistics before terminating.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Table 5: Effect of the beam threshold on the number of search errors (147 sentences).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We show the results of three of the experiments we conducted to measure isolated constituent precision under various partitioning schemes.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Again, famous place names will most likely be found in the dictionary, but less well-known names, such as 1PM\u00c2\u00b1 R; bu4lang3-shi4wei2-ke4 'Brunswick' (as in the New Jersey town name 'New Brunswick') will not generally be found.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The aforementioned surface form bcl, for example, may also stand for the lexical item \u201conion\u201d, a Noun.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The cost estimate, cost(i\u00c2\u00a5JJ1l.fn is computed in the obvious way by summing the negative log probabilities of i\u00c2\u00a5JJ1l.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Xim, } associated with the ith example.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Process nominals name the action of the transitive or ditransitive verb from which they derive.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"For \u00c3\u0086 = 0, no new target word is generated, while an additional source sentence position is covered.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"pre-processing.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The Recency KS computes the distance between the candidate and the anaphor relative to its scope.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Each out-of-domain phrase pair is characterized by a set of simple features intended to reflect how useful it will be.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We propose a limit of 70 words for Arabic parsing evaluations.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The parser switching oracle is the upper bound on the accuracy that can be achieved on this set in the parser switching framework.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The method shares some characteristics of the decision list algorithm presented in this paper.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"All commentaries have been annotated with rhetorical structure, using RSTTool4 and the definitions of discourse relations provided by Rhetorical Structure Theory (Mann, Thompson 1988).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The PROBING model was designed to improve upon SRILM by using linear probing hash tables (though not arranged in a trie), allocating memory all at once (eliminating the need for full pointers), and being easy to compile.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Clearly the percentage of productively formed words is quite small (for this particular corpus), meaning that dictionary entries are covering most of the 15 GR is .73 or 96%..","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We now describe the CoBoost algorithm for the named entity problem.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Here we push the single-framework conjecture across the board and present a single model that performs morphological segmentation and syntactic disambiguation in a fully generative framework.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The zone to which a token belongs is used as a feature.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For example, in Information Retrieval (IR), we have to match a user\u00e2\u0080\u0099s query to the expressions in the desired documents, while in Question Answering (QA), we have to find the answer to the user\u00e2\u0080\u0099s question even if the formulation of the answer in the document is different from the question.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Global features are extracted from other occurrences of the same token in the whole document.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Annotation of syntactic structure for the core corpus has just begun.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Based on these high-confidence alignments we can extract tuples of the form [u H v], where u is a foreign trigram type, whose middle word aligns to an English word type v. Our bilingual similarity function then sets the edge weights in proportion to these tuple counts.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The state-to-tag mapping is obtained from the best hyperparameter setting for 11 mapping shown in Table 3.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The ratio of buckets to entries is controlled by space multiplier m > 1.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Nicola Bertoldi and Marcello Federico assisted with IRSTLM.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In order to solve this problem, a parse tree is needed to understand that \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is not the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Of course, we are primarily interested in applying our techniques to languages for which no labeled resources are available.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"2.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Two subjects are each given a calendar and they are asked to schedule a meeting.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(b) 89 :1 t& tal de cai2neng2 hen3 he DE talent very 'He has great talent' f.b ga ol hig h While the current algorithm correctly handles the (b) sentences, it fails to handle the (a) sentences, since it does not have enough information to know not to group the sequences.ma3lu4 and?]cai2neng2 respectively.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"A generative model was applied (similar to naive Bayes) with the three labels as hidden vanables on unlabeled examples, and observed variables on (seed) labeled examples.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Specifically, the lexicon is generated as: P (T , W |\u00cf\u0088) =P (T )P (W |T ) Word Type Features (FEATS): Past unsupervised POS work have derived benefits from features on word types, such as suffix and capitalization features (Hasan and Ng, 2009; Berg-Kirkpatrick et al.,2010).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Recently, this topic has been getting more attention, as is evident from the Paraphrase Workshops in 2003 and 2004, driven by the needs of various NLP applications.","label":1},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The rationale for treating these semantic labels differently is that they are specific and reliable (as opposed to the WordNet classes, which are more coarse and more noisy due to polysemy).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"In contrast, NNP (proper nouns) form a large portion of vocabulary.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Some approaches depend upon some form of con\u00c2\u00ad straint satisfaction based on syntactic or semantic features (e.g., Yeh and Lee [1991], which uses a unification-based approach).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The human evaluators were a non-native, fluent Arabic speaker (the first author) for the ATB and a native English speaker for the WSJ.7 Table 5 shows type- and token-level error rates for each corpus.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For example, suppose one is building a ITS system for Mandarin Chinese.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"As we have seen, the lexicon of basic words and stems is represented as a WFST; most arcs in this WFST represent mappings between hanzi and pronunciations, and are costless.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"One is that smaller sets sometime have meaningless keywords, like \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d in the CC-domain, or \u00e2\u0080\u009ccompare\u00e2\u0080\u009d in the PC-domain.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Second, BABAR performs reliable case resolution to identify anaphora that can be easily resolved using the lexical and syntactic heuristics described in Section 2.1.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"A search restriction especially useful for the translation direction from German to English is presented.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The experimental tests are carried out on the Verbmobil task (GermanEnglish, 8000-word vocabulary), which is a limited-domain spoken-language task.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"i=1 (f,v)\u00e2\u0088\u0088Wi","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In considering this aspect of a formalism, we hope to better understand the relationship between the structural descriptions generated by the grammars of a formalism, and the properties of semilinearity and polynomial recognizability.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Less frequently studied is the interplay among language, annotation choices, and parsing model design (Levy and Manning, 2003; Ku\u00c2\u00a8 bler, 2005).","label":0},{"source_text":"There is no global pruning.","target_text":"The model is often further restricted so that each source word is assigned to exactly one target word (Brown et al., 1993; Ney et al., 2000).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"They cluster NE instance pairs based on the words in the contexts using a bag- of-words method.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"If the same pair of NE instances is used with different phrases, these phrases are likely to be paraphrases.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"12 For English, our Evalb implementation is identical to the most recent reference (EVALB20080701).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Without using the same test corpus, direct comparison is obviously difficult; fortunately, Chang et al. include a list of about 60 sentence fragments that exemplify various categories of performance for their system.","label":0},{"source_text":"There is no global pruning.","target_text":"Otherwise for the predecessor search hypothesis, we would have chosen a position that would not have been among the first n uncovered positions.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"These results are achieved by training on the official MUC6 and MUC7 training data, which is much less training data than is used by other machine learning systems that worked on the MUC6 or MUC7 named entity task (Bikel et al., 1997; Bikel et al., 1999; Borth- wick, 1999).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"3.2 Results.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The core of Yarowsky's algorithm is as follows: where h is defined by the formula in equation 2, with counts restricted to training data examples that have been labeled in step 2.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"On a set of 11 sentence fragments-the A set-where they reported 100% recall and precision for name identification, we had 73% recall and 80% precision.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The model we use provides a simple framework in which to incorporate a wide variety of lexical information in a uniform way.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Although we have not done any experiments on other languages, this way of using global features from a whole document should be applicable to other languages.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"For the TM, this is: where cI(s, t) is the count in the IN phrase table of pair (s, t), po(s|t) is its probability under the OUT TM, and cI(t) = &quot;s, cI(s', t).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Second, the reduced number of hidden variables and parameters dramatically speeds up learning and inference.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The first issue relates to the completeness of the base lexicon.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This is consistent with the nature of these two settings: log-linear combination, which effectively takes the intersection of IN and OUT, does relatively better on NIST, where the domains are broader and closer together.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"4.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Thus we are interested not in extraction, but actual generation from representations that may be developed to different degrees of granularity.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We have developed a general approach for combining parsers when preserving the entire structure of a parse tree is important.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"For instance, on Spanish, the absolute gap on median performance is 10%.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"This feature ft incorporates information from the smoothed graph and prunes hidden states that are inconsistent with the thresholded vector tx.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"2.1.1 Lexical Seeding It is generally not safe to assume that multiple occurrences of a noun phrase refer to the same entity.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"3.1 Lexicon Component.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The hyperparameters \u00ce\u00b1 and \u00ce\u00b2 represent the concentration parameters of the token- and type-level components of the model respectively.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Note that we check for , the word preceding the consecutive sequence of initCaps tokens, since person prefixes like Mr., Dr., etc are not part of person names, whereas corporate suffixes like Corp., Inc., etc are part of corporate names.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In principle, it would be possible to encode the exact position of the syntactic head in the label of the arc from the linear head, but this would give a potentially infinite set of arc labels and would make the training of the parser very hard.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Second, rather than relying on a division of the corpus into manually-assigned portions, we use features intended to capture the usefulness of each phrase pair.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Nicola Bertoldi and Marcello Federico assisted with IRSTLM.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The following two error criteria are used in our experiments: mWER: multi-reference WER: We use the Levenshtein distance between the automatic translation and several reference translations as a measure of the translation errors.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Having explained the various layers of annotation in PCC, we now turn to the question what all this might be good for.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.3 Assigning Evidence Values.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Clearly, for judges h and h taking h as standard and computing the precision and recall for Jz yields the same results as taking h as the standard, and computing for h, 14 All evaluation materials, with the exception of those used for evaluating personal names were drawn.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Parser 3, the most accurate parser, was chosen 71% of the time, and Parser 1, the least accurate parser was chosen 16% of the time.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"First, a non-anaphoric NP classifier identifies definite noun phrases that are existential, using both syntactic rules and our learned existential NP recognizer (Bean and Riloff, 1999), and removes them from the resolution process.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Each ht is a function that predicts a label (+1 or \u20141) on examples containing a particular feature xt, while abstaining on other examples: The prediction of the strong hypothesis can then be written as We now briefly describe how to choose ht and at at each iteration.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Each feature group can be made up of many binary features.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"9 61.0 44.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Our clue is the NE instance pairs.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The pseudo-code describing the algorithm is given in Fig.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"May 1995).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Its only purpose is 3 This follows since each \u00ce\u00b8t has St \u00e2\u0088\u0092 1 parameters and.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"A sentence was withheld from section 22 because its extreme length was troublesome for a couple of the parsers.'","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"From the definition of TAG's, it follows that the choice of adjunction is not dependent on the history of the derivation.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"All commentaries have been tagged with part-of-speech information using Brants\u00e2\u0080\u0099 TnT1 tagger and the Stuttgart\/Tu\u00c2\u00a8bingen Tag Set automatic analysis was responsible for this decision.)","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Our System Wang, Li, and Chang a. 1\\!f!IP Eflltii \/1\\!f!J:P $1til I b. agm: I a m: c. 5 Bf is Bf 1 d. \"*:t: w _t ff 1 \"* :t: w_tff 1 g., , Transliteration\/Translation chen2zhongl-shenl qu3 'music by Chen Zhongshen ' huang2rong2 youlyoul de dao4 'Huang Rong said soberly' zhangl qun2 Zhang Qun xian4zhang3 you2qingl shang4ren2 hou4 'after the county president You Qing had assumed the position' lin2 quan2 'Lin Quan' wang2jian4 'Wang Jian' oulyang2-ke4 'Ouyang Ke' yinl qi2 bu4 ke2neng2 rong2xu3 tai2du2 er2 'because it cannot permit Taiwan Independence so' silfa3-yuan4zhang3 lin2yang2-gang3 'president of the Judicial Yuan, Lin Yanggang' lin2zhangl-hu2 jiangl zuo4 xian4chang3 jie3shuol 'Lin Zhanghu will give an ex\u00c2\u00ad planation live' jin4\/iang3 nian2 nei4 sa3 xia4 de jinlqian2 hui4 ting2zhi3 'in two years the distributed money will stop' gaoltangl da4chi2 ye1zi0 fen3 'chicken stock, a tablespoon of coconut flakes' you2qingl ru4zhu3 xian4fu3 lwu4 'after You Qing headed the county government' Table 5 Performance on morphological analysis.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Two sets of examples from Gan are given in (1) and (2) (:::::: Gan's Appendix B, exx.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We can now compare this algorithm to that of (Yarowsky 95).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"These two properties of the tree sets are not only linguistically relevant, but also have computational importance.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We refer to this process as Reliable Case Resolution because it involves finding cases of anaphora that can be easily resolved with their antecedents.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"4 Evaluation Results.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Specifically, the limit n starts at 5 and increases by 5 at each iteration.)","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"gaolxing4 'happy' => F.i'JF.i'J Jl!","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We hope that this will allow practitioners to apply our approach directly to languages for which no resources are available.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"How should the absence of vowels and syntactic markers influence annotation choices and grammar development?","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Graph construction does not require any labeled data, but makes use of two similarity functions.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For irt the Good-Turing estimate just discussed gives us an estimate of p(unseen(f,) I f,)-the probability of observing a previously unseen instance of a construction in ft given that we know that we have a construction in f,.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"To lower the barrier of entrance to the competition, we provided a complete baseline MT system, along with data resources.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Evaluation results within sets Table 1 shows the evaluation result based on the number of phrases in a set.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The reference medicine for Silapo is EPREX\/ERYPO, which contains epoetin alfa.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"i=1 (f,v)\u00e2\u0088\u0088Wi","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In addition, there are several approaches to non-projective dependency parsing that are still to be evaluated in the large (Covington, 1990; Kahane et al., 1998; Duchier and Debusmann, 2001; Holan et al., 2001; Hellwig, 2003).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Proper names are assumed to be coreferent if they match exactly, or if they closely match based on a few heuristics.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In any event, to date, we have not compared different methods for deriving the set of initial frequency estimates.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The interdependence between fb or 1\/!i, and 5:2 is not captured by our model, but this could easily be remedied.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Memory mapping also allows the same model to be shared across processes on the same machine.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Instead, we condition on the type-level tag assignments T . Specifically, let St = {i|Ti = t} denote the indices of theword types which have been assigned tag t accord ing to the tag assignments T . Then \u00ce\u00b8t is drawn from DIRICHLET(\u00ce\u00b1, St), a symmetric Dirichlet which only places mass on word types indicated by St. This ensures that each word will only be assigned a single tag at inference time (see Section 4).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Because there might be some controversy about the exact definitions of such universals, this set of coarse-grained POS categories is defined operationally, by collapsing language (or treebank) specific distinctions to a set of categories that exists across all languages.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"This enables us to build a high performance NER without using separate classifiers to take care of global consistency or complex formulation on smoothing and backoff models (Bikel et al., 1997).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"3.2 Inter-annotator Agreement.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"more frequently than is done in English.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Part-of-speech (POS) tag distributions are known to exhibit sparsity \u00e2\u0080\u0094 a word is likely to take a single predominant tag in a corpus.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Unsupervised Learning of Contextual Role Knowledge for Coreference Resolution","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The Bikel GoldPOS configuration only supplies the gold POS tags; it does not force the parser to use them.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Even without features, but still using the tag prior, our median result is 52.0%, still significantly outperforming Grac\u00c2\u00b8a et al.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Different sentence structure and rich target language morphology are two reasons for this.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The most frequently occurring last words of organization names in cslist are compiled into a list of corporate suffixes, Corporate-Suffix- List.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The implementation is In a similar effort, (G\u00c2\u00a8otze 2003) developed a proposal for the theory-neutral annotation of information structure (IS) \u00e2\u0080\u0094 a notoriously difficult area with plenty of conflicting and overlapping terminological conceptions.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The search starts in the hypothesis (I; f;g; 0).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Minimal perfect hashing is used to find the index at which a quantized probability and possibly backoff are stored.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We do not attempt to identify the types of relationships that are found.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":".., Wn ) (obs) P T : Tag assigns (T1 ,.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The NE tagger is a rule-based system with 140 NE categories [Sekine et al. 2004].","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Based on revision 4041, we modified Moses to print process statistics before terminating.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The combining algorithm is presented with the candidate parses and asked to choose which one is best.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This process is repeated 5 times by rotating the data appropriately.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"In Table 1 we see with very few exceptions that the isolated constituent precision is less than 0.5 when we use the constituent label as a feature.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For example, in Information Retrieval (IR), we have to match a user\u00e2\u0080\u0099s query to the expressions in the desired documents, while in Question Answering (QA), we have to find the answer to the user\u00e2\u0080\u0099s question even if the formulation of the answer in the document is different from the question.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"It is difficult to motivate these days why one ministry should be exempt from cutbacks \u00e2\u0080\u0094 at the expense of the others.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Each xt E 2x is the set of features constituting the ith example.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The P (W |T , \u00cf\u0088) term in the lexicon component now decomposes as: n P (W |T , \u00cf\u0088) = n P (Wi|Ti, \u00cf\u0088) i=1 n \u00ef\u00a3\u00ab \u00ef\u00a3\u00b6 tions are not modeled by the standard HMM, which = n \u00ef\u00a3\u00ad n P (v|\u00cf\u0088Ti f )\u00ef\u00a3\u00b8 instead can model token-level frequency.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We then extract a set of possible tags tx(y) by eliminating labels whose probability is below a threshold value \u03c4: We describe how we choose \u03c4 in \u00a76.4.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"As in the case of the derivation trees of CFG's, nodes are labeled by a member of some finite set of symbols (perhaps only implicit in the grammar as in TAG's) used to denote derived structures.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The Penn Arabic Treebank (ATB) syntactic guidelines (Maamouri et al., 2004) were purposefully borrowed without major modification from English (Marcus et al., 1993).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This smooth guarantees that there are no zeroes estimated.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The monotone search performs worst in terms of both error rates mWER and SSER.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"As can be seen, GR and this \"pared-down\" statistical method perform quite similarly, though the statistical method is still slightly better.16 AG clearly performs much less like humans than these methods, whereas the full statistical algorithm, including morphological derivatives and names, performs most closely to humans among the automatic methods.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Jud ges A G G R ST M 1 M 2 M 3 T1 T2 T3 AG 0.7 0 0.7 0 0 . 4 3 0.4 2 0.6 0 0.6 0 0.6 2 0.5 9 GR 0.9 9 0 . 6 2 0.6 4 0.7 9 0.8 2 0.8 1 0.7 2 ST 0 . 6 4 0.6 7 0.8 0 0.8 4 0.8 2 0.7 4 M1 0.7 7 0.6 9 0.7 1 0.6 9 0.7 0 M2 0.7 2 0.7 3 0.7 1 0.7 0 M3 0.8 9 0.8 7 0.8 0 T1 0.8 8 0.8 2 T2 0.7 8 respectively, the recall and precision.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Text generation, or at least the two phases of text planning and sentence planning, is a process driven partly by well-motivated choices (e.g., use this lexeme X rather than that more colloquial near-synonym Y ) and partly by con tation like that of PCC can be exploited to look for correlations in particular between syntactic structure, choice of referring expressions, and sentence-internal information structure.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Find keywords for each NE pair The keywords are found for each NE category pair.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The goal of our research was to explore the use of contextual role knowledge for coreference resolution.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"To lower the barrier of entrance to the competition, we provided a complete baseline MT system, along with data resources.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For each domain, we created a blind test set by manually annotating 40 doc uments with anaphoric chains, which represent sets of m3 (S) = ) X \u00e2\u0088\u00a9Y =S 1 \u00e2\u0088\u0092 ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) (1) noun phrases that are coreferent (as done for MUC6 (MUC6 Proceedings, 1995)).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The most popular approach to dealing with seg\u00c2\u00ad mentation ambiguities is the maximum matching method, possibly augmented with further heuristics.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The procedure using the tagged sentences to discover paraphrases takes about one hour on a 2GHz Pentium 4 PC with 1GB of memory.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Given names are most commonly two hanzi long, occasionally one hanzi long: there are thus four possible name types, which can be described by a simple set of context-free rewrite rules such as the following: 1.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"For the LM, adaptive weights are set as follows: where \u03b1 is a weight vector containing an element \u03b1i for each domain (just IN and OUT in our case), pi are the corresponding domain-specific models, and \u02dcp(w, h) is an empirical distribution from a targetlanguage training corpus\u2014we used the IN dev set for this.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Both implementations employ a state object, opaque to the application, that carries information from one query to the next; we discuss both further in Section 4.2.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"A non-optimal analysis is shown with dotted lines in the bottom frame.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Each of the constituents must have received at least 1 votes from the k parsers, so a > I1 and 2 \u2014 2k\u00b1-1 b > ri-5-111.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Raj and Whittaker (2003) show that integers in a trie implementation can be compressed substantially.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The use of the Good-Turing equation presumes suitable estimates of the unknown expectations it requires.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Schapire and Singer show that the training error is bounded above by Thus, in order to greedily minimize an upper bound on training error, on each iteration we should search for the weak hypothesis ht and the weight at that minimize Z.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"VBD she added VP PUNC \u00e2\u0080\u009c SBAR IN NP 0 NN.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Limitations There are several limitations in the methods.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Step 2.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"This algorithm can be applied to statistical machine translation.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Methods for expanding the dictionary include, of course, morphological rules, rules for segmenting personal names, as well as numeral sequences, expressions for dates, and so forth (Chen and Liu 1992; Wang, Li, and Chang 1992; Chang and Chen 1993; Nie, Jin, and Hannan 1994).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"This modification brings the method closer to the DL-CoTrain algorithm described earlier, and is motivated by the intuition that all three labels should be kept healthily populated in the unlabeled examples, preventing one label from dominating \u2014 this deserves more theoretical investigation.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"How should the absence of vowels and syntactic markers influence annotation choices and grammar development?","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The performance of our system on those sentences ap\u00c2\u00ad peared rather better than theirs.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The composition operations in the case of CFG's are parameterized by the productions.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In our situation, the competing hypotheses are the possible antecedents for an anaphor.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Particular relations are also consistent with particular hypotheses about the segmentation of a given sentence, and the scores for particular relations can be incremented or decremented depending upon whether the segmentations with which they are consistent are \"popular\" or not.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"At each iteration the algorithm increases the number of rules, while maintaining a high level of agreement between the spelling and contextual decision lists.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Besides size of training data, the use of dictionaries is another factor that might affect performance.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"4.3 Morphological Analysis.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"6One of our experimental settings lacks document boundaries, and we used this approximation in both settings for consistency.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The only way to handle such phenomena within the framework described here is simply to expand out the reduplicated forms beforehand, and incorporate the expanded forms into the lexical transducer.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"IRST is not threadsafe.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"At the same time, the n-gram error rate is sensitive to samples with extreme n-gram counts.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The 1-bit sign is almost always negative and the 8-bit exponent is not fully used on the range of values, so in practice this corresponds to quantization ranging from 17 to 20 total bits.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"However, there are several reasons why this approach will not in general work: 1.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"While Gan's system incorporates fairly sophisticated models of various linguistic information, it has the drawback that it has only been tested with a very small lexicon (a few hundred words) and on a very small test set (thirty sentences); there is therefore serious concern as to whether the methods that he discusses are scalable.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"besuchen 9.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This Good\u00c2\u00ad Turing estimate of p(unseen(f,) If,) can then be used in the normal way to define the probability of finding a novel instance of a construction in ir, in a text: p(unseen(f,)) = p(unseen(f,) I f,) p(fn Here p(ir,) is just the probability of any construction in ft as estimated from the frequency of such constructions in the corpus.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"For effectively annotating connectives\/scopes, we found that existing annotation tools were not well-suited, for two reasons: \u00e2\u0080\u00a2 Some tools are dedicated to modes of annotation (e.g., tiers), which could only quite un-intuitively be used for connectives and scopes.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Limitations of (Blum and Mitchell 98): While the assumptions of (Blum and Mitchell 98) are useful in developing both theoretical results and an intuition for the problem, the assumptions are quite limited.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Instead, we condition on the type-level tag assignments T . Specifically, let St = {i|Ti = t} denote the indices of theword types which have been assigned tag t accord ing to the tag assignments T . Then \u00ce\u00b8t is drawn from DIRICHLET(\u00ce\u00b1, St), a symmetric Dirichlet which only places mass on word types indicated by St. This ensures that each word will only be assigned a single tag at inference time (see Section 4).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"2 62.6 45.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Table 3 shows BABAR\u00e2\u0080\u0099s performance when the four contextual role knowledge sources are added.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"To date we have not done a separate evaluation of foreign-name recognition.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The first row represents the average accuracy of the three parsers we combine.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We then show that the combining techniques presented above give better parsing accuracy than any of the individual parsers.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The effect of UNIQ is very small on both data sets.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"But diacritics are not present in unvocalized text, which is the standard form of, e.g., news media documents.3 VBD she added VP PUNC S VP VBP NP ...","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"This approach makes the training objective more complex by adding linear constraints proportional to the number of word types, which is rather prohibitive.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"(Charniak et al., 1996).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The function A : F \u2014* C maps from the language specific fine-grained tagset F to the coarser universal tagset C and is described in detail in \u00a76.2: Note that when tx(y) = 1 the feature value is 0 and has no effect on the model, while its value is \u2212oc when tx(y) = 0 and constrains the HMM\u2019s state space.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The model was built with open vocabulary, modified Kneser-Ney smoothing, and default pruning settings that remove singletons of order 3 and higher.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The advantage is that we can recombine search hypotheses by dynamic programming.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The dev corpus was taken from the NIST05 evaluation set, augmented with some randomly-selected material reserved from the training set.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The compressed variant uses block compression and is rather slow as a result.","label":0},{"source_text":"There is no global pruning.","target_text":"To be short, we omit the target words e; e0 in the formulation of the search hypotheses.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Table 1 shows results of the benchmark.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"However, for our purposes it is not sufficient to repre\u00c2\u00ad sent the morphological decomposition of, say, plural nouns: we also need an estimate of the cost of the resulting word.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"2 62.6 45.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The present proposal falls into the last group.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"(Levinger et al., 1995; Goldberg et al., ; Adler et al., 2008)) will make the parser more robust and suitable for use in more realistic scenarios.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"For instance, the sentence Similar improvements in haemoglobin levels were reported in the scientific literature for other epoetins would likely be considered domain-specific despite the presence of general phrases like were reported in.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Two aspects of the corpus have been presented in previous papers ((Re- itter, Stede 2003) on underspecified rhetorical structure; (Stede 2003) on the perspective of knowledge-based summarization).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The corresponding token words w are drawn conditioned on t and \u00ce\u00b8.2 Our full generative model is given by: K P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) = n (P (\u00cf\u0086t|\u00ce\u00b1)P (\u00ce\u00b8t|T , \u00ce\u00b1)) t=1 The transition distribution \u00cf\u0086t for each tag t is drawn according to DIRICHLET(\u00ce\u00b1, K ), where \u00ce\u00b1 is the shared transition and emission distribution hyperparameter.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Various verbal (e.g., \u00ef\u00bf\u00bd, .::) and adjectival.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We use the universal POS tagset of Petrov et al. (2011) in our experiments.10 This set C consists of the following 12 coarse-grained tags: NOUN (nouns), VERB (verbs), ADJ (adjectives), ADV (adverbs), PRON (pronouns), DET (determiners), ADP (prepositions or postpositions), NUM (numerals), CONJ (conjunctions), PRT (particles), PUNC (punctuation marks) and X (a catch-all for other categories such as abbreviations or foreign words).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"2 62.2 39.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The maximum likelihood estimates (i.e., parameter values which maximize 10) can not be found analytically, but the EM algorithm can be used to hill-climb to a local maximum of the likelihood function from some initial parameter settings.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"CHARACTERIZING STRUCTURAL DESCRIPTIONS PRODUCED BY VARIOUS GRAMMATICAL FORMALISMS*","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The NP is a complement to a preposition, which is the head of a PP.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Manual annotation results in human in- terpretable grammars that can inform future tree- bank annotation decisions.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"gaolgaolxing4xing4 'happily' In the particular form of A-not-A reduplication illustrated in (3a), the first syllable of the verb is copied, and the negative markerbu4 'not' is inserted between the copy and the full verb.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"(1998) did make use of information from the whole document.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"In both cases, the instanceweighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline, and gains of between 0.6 and 1.8 over an equivalent mixture model (with an identical training procedure but without instance weighting).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"87 Table 7: Test set results.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(We would like to note though that unlike previous boosting algorithms, the CoBoost algorithm presented here is not a boosting algorithm under Valiant's (Valiant 84) Probably Approximately Correct (PAC) model.)","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The possible analyses of a surface token pose constraints on the analyses of specific segments.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Out of those 15 links, 4 are errors, namely \u00e2\u0080\u009cbuy - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cpurchase - stake\u00e2\u0080\u009d \u00e2\u0080\u009cacquisition - stake\u00e2\u0080\u009d.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"A position is presented by the word at that position.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Some relations are signalled by subordinating conjunctions, which clearly demarcate the range of the text spans related (matrix clause, embedded clause).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"State is implemented in their scrolling variant, which is a trie annotated with forward and backward pointers.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"This approach needs a phrase as an initial seed and thus the possible relationships to be extracted are naturally limited.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"(Thus the domain of the dev and test corpora matches IN.)","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(Blum and Mitchell 98) give an example that illustrates just how powerful the second constraint can be.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Other kinds of productive word classes, such as company names, abbreviations (termed fijsuolxie3 in Mandarin), and place names can easily be 20 Note that 7 in E 7 is normally pronounced as leO, but as part of a resultative it is liao3..","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The string pumping lemma for CFG's (uvwxy-theorem) can be seen as a corollary of this lemma. from this pumping lemma: a single path can be pumped independently.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Borth 2MUC data can be obtained from the Linguistic Data Consortium: http:\/\/www.ldc.upenn.edu 3Training data for IdentiFinder is actually given in words (i.e., 650K & 790K words), rather than tokens Table 6: Comparison of results for MUC7 wick (1999) reported using dictionaries of person first names, corporate names and suffixes, colleges and universities, dates and times, state abbreviations, and world regions.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In these examples, the names identified by the two systems (if any) are underlined; the sentence with the correct segmentation is boxed.19 The differences in performance between the two systems relate directly to three issues, which can be seen as differences in the tuning of the models, rather than repre\u00c2\u00ad senting differences in the capabilities of the model per se.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The Verbmobil task is an appointment scheduling task.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"This work was supported in part by the Swedish Research Council (621-2002-4207).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Inflectional features marking pronominal elements may be attached to different kinds of categories marking their pronominal complements.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Supervised methods have been applied quite successfully to the full MUC named-entity task (Bikel et al. 97).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Supervised learning approaches have advanced the state-of-the-art on a variety of tasks in natural language processing, resulting in highly accurate systems.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"This is the first set that gives us a fair evaluation of the Bayes models, and the Bayes switching model performs significantly better than its non-parametric counterpart.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"More complex approaches such as the relaxation technique have been applied to this problem Fan and Tsai (1988}.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Dictionaries: Due to the limited amount of training material, name dictionaries have been found to be useful in the named entity task.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Many packages perform language model queries.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"As was explained in the results section, \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d are not desirable keywords in the CC-domain.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Unlike Germann et al. (2009), we chose a model size so that all benchmarks fit comfortably in main memory.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"3 These are not full case frames in the traditional sense, but they approximate a simple case frame with a single slot.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Therefore, we want state to encode the minimum amount of information necessary to properly compute language model scores, so that the decoder will be faster and make fewer search errors.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"3.1 Corpora.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"MENE is then trained on 80% of the training corpus, and tested on the remaining 20%.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The combining technique must act as a multi-position switch indicating which parser should be trusted for the particular sentence.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Most similar is scrolling queries, wherein left-to-right queries that add one word at a time are optimized.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"i=1 (f,v)\u00e2\u0088\u0088Wi","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"If it is made up of all capital letters, then (allCaps, zone) is set to 1.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The data structure was populated with 64-bit integers sampled uniformly without replacement.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"One key issue here is to seek a discourse-based model of information structure.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We again adopt an approach where we alternate between two classifiers: one classifier is modified while the other remains fixed.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"), and thosethat begin with a verb (\u00ef\u00bf\u00bd ub..i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"If the expression is a word or a short phrase (like \u00e2\u0080\u009ccorporation\u00e2\u0080\u009d and \u00e2\u0080\u009ccompany\u00e2\u0080\u009d), it is called a \u00e2\u0080\u009csynonym\u00e2\u0080\u009d.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The final model tions.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Our experiments consistently demonstrate that this model architecture yields substantial performance gains over more complex tagging counterparts.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"While building a machine translation system is a serious undertaking, in future we hope to attract more newcomers to the field by keeping the barrier of entry as low as possible.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Section 4.1 explained that state s is stored by applications with partial hypotheses to determine when they can be recombined.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"07 80.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Subsets of partial hypotheses with coverage sets C of increasing cardinality c are processed.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Unfortunately, we were unable to correctly run the IRSTLM quantized variant.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The web-based Annis imports data in a variety of XML formats and tagsets and displays it in a tier-orientedway (optionally, trees can be drawn more ele gantly in a separate window).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For E(ni1s), then, we substitute a smooth S against the number of class elements.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"of Articles No.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"As suggested in Section 4.3.2, a derivation with independent paths can be divided into subcomputations with limited sharing of information.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Moving beyond directly related work, major themes in SMT adaptation include the IR (Hildebrand et al., 2005; L\u00a8u et al., 2007; Zhao et al., 2004) and mixture (Finch and Sumita, 2008; Foster and Kuhn, 2007; Koehn and Schroeder, 2007; L\u00a8u et al., 2007) approaches for LMs and TMs described above, as well as methods for exploiting monolingual in-domain text, typically by translating it automatically and then performing self training (Bertoldi and Federico, 2009; Ueffing et al., 2007; Schwenk and Senellart, 2009).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"+ cost(unseen(fm, as desired.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Figure 4 shows some such phrase sets based on keywords in the CC-domain.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"As the reviewer also points out, this is a problem that is shared by, e.g., probabilistic context-free parsers, which tend to pick trees with fewer nodes.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The performance of our system on those sentences ap\u00c2\u00ad peared rather better than theirs.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"In speech recognition the arcs of the lattice are typically weighted in order to indicate the probability of specific transitions.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Out of those 15 links, 4 are errors, namely \u00e2\u0080\u009cbuy - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cpurchase - stake\u00e2\u0080\u009d \u00e2\u0080\u009cacquisition - stake\u00e2\u0080\u009d.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We use the HSPELL9 (Har\u2019el and Kenigsberg, 2004) wordlist as a lexeme-based lexicon for pruning segmentations involving invalid segments.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Finally, Section 5 explains how BABAR relates to previous work, and Section 6 summarizes our conclusions.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Floating point values may be stored in the trie exactly, using 31 bits for non-positive log probability and 32 bits for backoff5.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Linear weights are difficult to incorporate into the standard MERT procedure because they are \u201chidden\u201d within a top-level probability that represents the linear combination.1 Following previous work (Foster and Kuhn, 2007), we circumvent this problem by choosing weights to optimize corpus loglikelihood, which is roughly speaking the training criterion used by the LM and TM themselves.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"07 95.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Table 2 shows results for both settings and all methods described in sections 2 and 3.","label":0},{"source_text":"There is no global pruning.","target_text":"3.1 Word ReOrdering with Verbgroup.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"07 80.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"As noted in Section 1, our code finds the longest matching entry wnf for query p(wn|s(wn\u22121 f ) The probability p(wn|wn\u22121 f ) is stored with wnf and the backoffs are immediately accessible in the provided state s(wn\u22121 When our code walks the data structure to find wnf , it visits wnn, wnn\u22121, ... , wnf .","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"vierten 12.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"However, the next step is clearly different.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Search CPU time mWER SSER Method [sec] [%] [%] MonS 0:9 42:0 30:5 QmS 10:6 34:4 23:8 IbmS 28:6 38:2 26:2 4.2 Performance Measures.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The difference is due to more careful S-NOM NP NP NP VP VBG :: b NP restoring NP ADJP NN :: b NP NN NP NP ADJP DTJJ ADJP DTJJ NN :: b NP NP NP ADJP ADJP DTJJ J ..i NN :: b NP NP NP ADJP ADJP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ DTJJ J ..i _;\u00ef\u00bf\u00bd PRP J ..i NN _;\u00ef\u00bf\u00bd NP PRP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ J ..i role its constructive effective (b) Stanford (c) Berkeley (d) Bik el (a) Reference Figure 4: The constituent Restoring of its constructive and effective role parsed by the three different models (gold segmentation).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"While size of the resulting transducers may seem daunting-the segmenter described here, as it is used in the Bell Labs Mandarin TTS system has about 32,000 states and 209,000 arcs-recent work on minimization of weighted machines and transducers (cf.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Two general approaches are presented and two combination techniques are described for each approach.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"16 As one reviewer points out, one problem with the unigram model chosen here is that there is still a. tendency to pick a segmentation containing fewer words.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"A large number of rules is needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classi- However, we show that the use of data can reduce the requirements for supervision to just 7 simple &quot;seed&quot; rules.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"set was based on an earlier version of the Chang et a!.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Trees derived by IG's exhibit a property that is not exhibited by the trees sets derived by TAG's or CFG's.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"more frequently than is done in English.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"M must enter a universal state and check that each of the k constant substrings are in the appropriate place (as determined by the contents of the first 2k work tapes) on the input tape.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"2 70.7 52.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"To make the projection practical, we rely on the twelve universal part-of-speech tags of Petrov et al. (2011).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Further, the special hash 0 suffices to flag empty buckets.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This feature imposes constraints Table 1: Features based on the token string that are based on the probability of each name class during training.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Compared with SRILM, IRSTLM adds several features: lower memory consumption, a binary file format with memory mapping, caching to increase speed, and quantization.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"All the sentences have been analyzed by our chunker and NE tag- ger.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"This process is repeated 5 times by rotating the data appropriately.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Nonetheless, parse quality is much lower in the joint model because a lattice is effectively a long sentence.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This makes memory usage comparable to our PROBING model.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"TIS systems in general need to do more than simply compute the.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"If the log backoff of wnf is also zero (it may not be in filtered models), then wf should be omitted from the state.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(4) gives In order to minimize Zt, at each iteration the final algorithm should choose the weak hypothesis (i.e., a feature xt) which has values for W+ and W_ that minimize Equ.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Morphological segmentation decisions in our model are delegated to a lexeme-based PCFG and we show that using a simple treebank grammar, a data-driven lexicon, and a linguistically motivated unknown-tokens handling our model outperforms (Tsarfaty, 2006) and (Cohen and Smith, 2007) on the joint task and achieves state-of-the-art results on a par with current respective standalone models.2","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","label":1},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The features are weighted within a logistic model to give an overall weight that is applied to the phrase pair\u2019s frequency prior to making MAP-smoothed relative-frequency estimates (different weights are learned for each conditioning direction).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"vierten 12.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The precision and recall of similarity switching and constituent voting are both significantly better than the best individual parser, and constituent voting is significantly better than parser switching in precision.4 Constituent voting gives the highest accuracy for parsing the Penn Treebank reported to date.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, the company may refer to Company X in one paragraph and Company Y in another.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Each trie node is individually allocated and full 64-bit pointers are used to find them, wasting memory.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Xim, } associated with the ith example.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Time for Moses itself to load, including loading the language model and phrase table, is included.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The precision and recall of similarity switching and constituent voting are both significantly better than the best individual parser, and constituent voting is significantly better than parser switching in precision.4 Constituent voting gives the highest accuracy for parsing the Penn Treebank reported to date.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"As a (crude) approximation, we normalize the extraction patterns with respect to active and passive voice and label those extractions as agents or patients.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"It should be clear from the onset that the particle b (\u201cin\u201d) in \u2018bcl\u2019 may then attach higher than the bare noun cl (\u201cshadow\u201d).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Linguistic intuitions like those in the previous section inform language-specific annotation choices.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"This feature has a linguistic justification.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"That is, we can use the discourse parser on PCC texts, emulating for instance a \u00e2\u0080\u009cco-reference oracle\u00e2\u0080\u009d that adds the information from our co-reference annotations.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We used 22 features for the logistic weighting model, divided into two groups: one intended to reflect the degree to which a phrase pair belongs to general language, and one intended to capture similarity to the IN domain.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In (b) is a plausible segmentation for this sentence; in (c) is an implausible segmentation.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"3.1 Corpora.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"For the translation model Pr(fJ 1 jeI 1), we go on the assumption that each source word is aligned to exactly one target word.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"In the label propagation stage, we propagate the automatic English tags to the aligned Italian trigram types, followed by further propagation solely among the Italian vertices. the Italian vertices are connected to an automatically labeled English vertex.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Instead, we extend the variation n-gram method of Dickinson (2005) to compare annotation error rates in the WSJ and ATB.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"To investigate the influence of these factors, we analyze Modern Standard Arabic (henceforth MSA, or simply \u00e2\u0080\u009cArabic\u00e2\u0080\u009d) because of the unusual opportunity it presents for comparison to English parsing results.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We will describe the evaluation of such clusters in the next subsection.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Table 2 shows the features used in the current version of the parser.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"A dynamic programming algorithm is then used to select the sequence of word classes with the highest probability.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"These systems rely on a training corpus that has been manually annotated with coreference links.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The 95% confidence intervals for type-level errors are (5580, 9440) for the ATB and (1400, 4610) for the WSJ.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The two new terms force the two classifiers to agree, as much as possible, on the unlabeled examples.","label":0},{"source_text":"There is no global pruning.","target_text":"German English Training: Sentences 58 073 Words 519 523 549 921 Words* 418 979 453 632 Vocabulary Size 7939 4648 Singletons 3454 1699 Test-147: Sentences 147 Words 1 968 2 173 Perplexity { 26:5 Table 4: Multi-reference word error rate (mWER) and subjective sentence error rate (SSER) for three different search procedures.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This paper, however, provides a comprehensive overview of the data collection effort and its current state.","label":0},{"source_text":"There is no global pruning.","target_text":"In the last example, the less restrictive IbmS word reordering leads to a better translation, although the QmS translation is still acceptable.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Yet, some hanzi are far more probable in women's names than they are in men's names, and there is a similar list of male-oriented hanzi: mixing hanzi from these two lists is generally less likely than would be predicted by the independence model.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A corollary of the result of Section 4.3 is that polynomial time recognition of MCTAG's is possible.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We follow the guidelines developed in the TIGER project (Brants et al. 2002) for syntactic annotation of German newspaper text, using the Annotate3 tool for interactive construction of tree structures.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The restriction can be expressed in terms of the number of uncovered source sentence positions to the left of the rightmost position m in the coverage set.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"These universal POS categories not only facilitate the transfer of POS information from one language to another, but also relieve us from using controversial evaluation metrics,2 by establishing a direct correspondence between the induced hidden states in the foreign language and the observed English labels.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Then, token- level HMM emission parameters are drawn conditioned on these assignments such that each word is only allowed probability mass on a single assigned tag.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"This may seem surprising, given the experiments reported in section 4, but the explanation is probably that the non-projective dependencies that can be recovered at all are of the simple kind that only requires a single lift, where the encoding of path information is often redundant.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The first unsupervised algorithm we describe is based on the decision list method from (Yarowsky 95).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE without Proteus, however, did not do very well and only achieved an F measure of 84.22% (Borthwick, 1999).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"However, it can be noted that the results for the least informative encoding, Path, are almost comparable, while the third encoding, Head, gives substantially worse results for both data sets.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This WFST represents the segmentation of the text into the words AB and CD, word boundaries being marked by arcs mapping between f and part-of-speech labels.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"As Figure 1 shows, this word has no high-confidence alignment in the Italian-English bitext.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"None of the parsers produce parses with crossing brackets, so none of them votes for both of the assumed constituents.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"La ng ua ge # To ke ns # W or d Ty pe s # Ta gs E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 1 1 7 3 7 6 6 9 4 3 8 6 2 0 3 5 6 8 6 9 9 6 0 5 2 0 6 6 7 8 8 9 3 3 4 1 9 1 4 6 7 4 9 2 0 6 1 8 3 5 6 2 8 3 9 3 7 2 3 2 5 2 8 9 3 1 1 6 4 5 8 2 0 0 5 7 4 5 2 5 1 2 5 4 2 2 4 7 4 1 Table 2: Statistics for various corpora utilized in experiments.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Given a sufficient number of randomly drawn unlabeled examples (i.e., edges), we will induce two completely connected components that together span the entire graph.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"There are clearly eight orthographic words in the example given, but if one were doing syntactic analysis one would probably want to consider I'm to consist of two syntactic words, namely I and am.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"63 81.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"A straightforward way to find the shortest tour is by trying all possible permutations of the n cities.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We therefore used the arithmetic mean of each interjudge precision-recall pair as a single measure of interjudge similarity.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"A straightforward way to find the shortest tour is by trying all possible permutations of the n cities.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Table 2 shows our complete set of results.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"3 Throughout this paper we shall give Chinese examples in traditional orthography, followed.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"It would have therefore also been possible to use the integer programming (IP) based approach of Ravi and Knight (2009) instead of the feature-HMM for POS induction on the foreign side.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The system will then look for sequences of initial capitalized words that match the acronyms found in the whole document.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"It is known that CFG's, HG's, and TAG's can be recognized in polynomial time since polynomial time algorithms exist in for each of these formalisms.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The dev corpus was taken from the NIST05 evaluation set, augmented with some randomly-selected material reserved from the training set.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(7), such as the likelihood function used in maximum-entropy problems and other generalized additive models (Lafferty 99).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The question is how to normalize the probabilities in such a way that smaller groupings have a better shot at winning.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Step 3.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Local features are features that are based on neighboring tokens, as well as the token itself.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Given the bilingual graph described in the previous section, we can use label propagation to project the English POS labels to the foreign language.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"For all languages, the vocabulary sizes increase by several thousand words.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"In addition to the optimizations specific to each datastructure described in Section 2, we implement several general optimizations for language modeling.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"As far as we know, no other NERs have used information from the whole document (global) as well as information within the same sentence (local) in one framework.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"37 79.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The difference in performance between pronouns and definite noun phrases surprised us.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Wu and Fung introduce an evaluation method they call nk-blind.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Lexicon Feature of Previous and Next Token: The string of the previous token and the next token is used with the initCaps information of . If has initCaps, then a feature (initCaps, ) is set to 1.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Because of their size, the examples (Figures 2 to 4) appear at the end of the paper.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The rest of the paper is structured as follows.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The sequence of states needed to carry out the word reordering example in Fig.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"At each point during the derivation, the prediction is based on six word tokens, the two topmost tokens on the stack, and the next four input tokens.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"An easy way to achieve this is to put the domain-specific LMs and TMs into the top-level log-linear model and learn optimal weights with MERT (Och, 2003).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"BABAR uses unsupervised learning to acquire this knowledge from plain text without the need for annotated training data.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"As can be seen in Table 4, our training data is a lot less than those used by MENE and IdentiFinder3.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"2.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"7 Conclusion and Future Work.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, there is again local grammatical information that should favor the split in the case of (1a): both .ma3 'horse' and .ma3 lu4 are nouns, but only .ma3 is consistent with the classifier pil, the classifier for horses.21 By a similar argument, the preference for not splitting , lm could be strengthened in (lb) by the observation that the classifier 1'1* tiao2 is consistent with long or winding objects like , lm ma3lu4 'road' but not with,ma3 'horse.'","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If they are found in a list, then a feature for that list will be set to 1.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Each xii is a member of X, where X is a set of possible features.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This has solutions: where pI(s|t) is derived from the IN corpus using relative-frequency estimates, and po(s|t) is an instance-weighted model derived from the OUT corpus.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"At each point during the derivation, the parser has a choice between pushing the next input token onto the stack \u2013 with or without adding an arc from the token on top of the stack to the token pushed \u2013 and popping a token from the stack \u2013 with or without adding an arc from the next input token to the token popped.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Full Chinese personal names are in one respect simple: they are always of the form family+given.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The weak hypothesis chosen was then restricted to be a predictor in favor of this label.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The PCFG was trained from the same sections of the Penn Treebank as the other three parsers.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Word frequencies are estimated by a re-estimation procedure that involves apply\u00c2\u00ad ing the segmentation algorithm presented here to a corpus of 20 million words,8 using 8 Our training corpus was drawn from a larger corpus of mixed-genre text consisting mostly of.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Microsoft\u2019s approach uses dependency trees, others use hierarchical phrase models.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Evaluation of links A link between two sets is considered correct if the majority of phrases in both sets have the same meaning, i.e. if the link indicates paraphrase.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"handled given appropriate models.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, suppose the current model assigns a belief value of .60 to {A, B}, meaning that it is 60% sure that the correct hypothesis is either A or B. Then new evidence arrives with a belief value of .70 assigned 5 Initially there are no competing hypotheses because all hypotheses are included in \u00ce\u00b8 by definition.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Also, the method of using keywords rules out phrases which don\u00e2\u0080\u0099t contain popular words in the domain.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This has led previous workers to adopt ad hoc linear weighting schemes (Finch and Sumita, 2008; Foster and Kuhn, 2007; L\u00a8u et al., 2007).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"In future work we plan to try this approach with more competitive SMT systems, and to extend instance weighting to other standard SMT components such as the LM, lexical phrase weights, and lexicalized distortion.","label":1},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The model was parameterized such that the joint probability of a (label, feature-set) pair P(yi, xi) is written as The model assumes that (y, x) pairs are generated by an underlying process where the label is first chosen with some prior probability P(yi); the number of features mi is then chosen with some probability P(mi); finally the features are independently generated with probabilities P(xulyi).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We believe that the underlying principles of the maximum entropy framework are suitable for exploiting information from diverse sources.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"At this stage the lattice path corresponds to segments only, with no PoS assigned to them.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"However, MADA is language-specific and relies on manually constructed dictionaries.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Zone: MUC data contains SGML tags, and a document is divided into zones (e.g., headlines and text zones).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We concentrate on those sets.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"2 61.7 64.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The overall performance of our joint framework demonstrates that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperforms upper bounds proposed by previous joint disambiguation systems and achieves segmentation and parsing results on a par with state-of-the-art standalone applications results.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"A receives a votes, and B receives b votes.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Model components cascade, so the row corresponding to +FEATS also includes the PRIOR component (see Section 3).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The corresponding token words w are drawn conditioned on t and \u00ce\u00b8.2 Our full generative model is given by: K P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) = n (P (\u00cf\u0086t|\u00ce\u00b1)P (\u00ce\u00b8t|T , \u00ce\u00b1)) t=1 The transition distribution \u00cf\u0086t for each tag t is drawn according to DIRICHLET(\u00ce\u00b1, K ), where \u00ce\u00b1 is the shared transition and emission distribution hyperparameter.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Moreover, they are used as substantives much 2 Unlike machine translation, constituency parsing is not significantly affected by variable word order.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In Section 2, we brie y review our approach to statistical machine translation.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We use w erations of sampling (see Figure 2 for a depiction).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This supports our main thesis that decisions taken by single, improved, grammar are beneficial for both tasks.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Schapire and Singer show that the training error is bounded above by Thus, in order to greedily minimize an upper bound on training error, on each iteration we should search for the weak hypothesis ht and the weight at that minimize Z.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"97 81.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Decreasing the threshold results in higher mWER due to additional search errors.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"A Stochastic Finite-State Word-Segmentation Algorithm for Chinese","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Process nominals name the action of the transitive or ditransitive verb from which they derive.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"For a sentence x and a state sequence z, a first order Markov model defines a distribution: (9) where Val(X) corresponds to the entire vocabulary.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The problem with these styles of evaluation is that, as we shall demonstrate, even human judges do not agree perfectly on how to segment a given text.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Overall, language modeling significantly impacts decoder performance.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"For example, Gazdar (1985) discusses the applicability of Indexed Grammars (IG's) to Natural Language in terms of the structural descriptions assigned; and Berwick (1984) discusses the strong generative capacity of Lexical-Functional Grammar (LFG) and Government and Bindings grammars (GB).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The reference medicine for Silapo is EPREX\/ERYPO, which contains epoetin alfa.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"In our case multi-threading is trivial because our data structures are read-only and uncached.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"And time is short.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"phrase (markContainsVerb).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"It also does not prune, so comparing to our pruned model would be unfair.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Various segmentation approaches were then compared with human performance: 1.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We call these N \u2212 1 words state.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Across eight European languages, our approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","label":1},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Two general approaches are presented and two combination techniques are described for each approach.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Let us notate the set of previously unseen, or novel, members of a category X as unseen(X); thus, novel members of the set of words derived in f, menO will be de\u00c2\u00ad noted unseen(f,).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Sometime, multiple words are needed, like \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, \u00e2\u0080\u009cprime minister\u00e2\u0080\u009d or \u00e2\u0080\u009cpay for\u00e2\u0080\u009d (\u00e2\u0080\u009cpay\u00e2\u0080\u009d and \u00e2\u0080\u009cpay for\u00e2\u0080\u009d are different senses in the CC-domain).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"1 61.2 43.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We outlined the definition of a family of constrained grammatical formalisms, called Linear Context-Free Rewriting Systems.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We do not adapt the alignment procedure for generating the phrase table from which the TM distributions are derived.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"\u00e2\u0080\u00a2 Bridging links: the annotator is asked to specify the type as part-whole, cause-effect (e.g., She had an accident.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"To explicitly handle the word reordering between words in source and target language, we use the concept of the so-called inverted alignments as given in (Ney et al., 2000).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Initially, the DempsterShafer model assumes that all hypotheses are equally likely, so it creates a set called \u00ce\u00b8 that includes all hypotheses.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Thus, the effects of spontaneous speech are present in the corpus, e.g. the syntactic structure of the sentence is rather less restricted, however the effect of speech recognition errors is not covered.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"2) An improved language model, which takes into account syntactic structure, e.g. to ensure that a proper English verbgroup is generated.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"This section describes an algorithm based on boosting algorithms, which were previously developed for supervised machine learning problems.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"An examination of the subjects' bracketings confirmed that these instructions were satisfactory in yielding plausible word-sized units.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Under certain conditions the constituent voting and na\u00efve Bayes constituent combination techniques are guaranteed to produce sets of constituents with no crossing brackets.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Finally, we add \u00e2\u0080\u009cDT\u00e2\u0080\u009d to the tags for definite nouns and adjectives (Kulick et al., 2006).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"97 78.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, even if the contexts surrounding an anaphor and candidate match exactly, they are not coreferent if they have substantially different meanings 9 We would be happy to make our manually annotated test data available to others who also want to evaluate their coreference resolver on the MUC4 or Reuters collections.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"NN \u00ef\u00bf\u00bd .e NP NNP NP DTNNP NN \u00ef\u00bf\u00bd .e NP NP NNP NP Table 5: Evaluation of 100 randomly sampled variation nuclei types.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We considered using the MUC6 and MUC7 data sets, but their training sets were far too small to learn reliable co-occurrence statistics for a large set of contextual role relationships.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"A greedy algorithm (or maximum-matching algorithm), GR: proceed through the sentence, taking the longest match with a dictionary entry at each point.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For example, the independence assumptions mean that the model fails to capture the dependence between specific and more general features (for example the fact that the feature full.-string=New_York is always seen with the features contains (New) and The baseline method tags all entities as the most frequent class type (organization). contains (York) and is never seen with a feature such as contains (Group) ).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"10 Here we use the Good-Turing estimate (Baayen 1989; Church and Gale 1991), whereby the aggregate probability of previously unseen instances of a construction is estimated as ni\/N, where N is the total number of observed tokens and n1 is the number of types observed only once.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"i=1 (f,v)\u00e2\u0088\u0088Wi","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Because there might be some controversy about the exact definitions of such universals, this set of coarse-grained POS categories is defined operationally, by collapsing language (or treebank) specific distinctions to a set of categories that exists across all languages.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For the moment we will assume that there are only two possible labels: each y, is in { \u20141, +1}.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"(1), Pr(eI 1) is the language model, which is a trigram language model in this case.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For RandLM and IRSTLM, the effect of caching can be seen on speed and memory usage.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In this section we present a partial evaluation of the current system, in three parts.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Employing a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions is not only theoretically clean and linguistically justified and but also probabilistically apropriate and empirically sound.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Exploiting Diversity in Natural Language Processing: Combining Parsers","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: The regime gives itself the right...","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Because the Bikel parser has been parameter- ized for Arabic by the LDC, we do not change the default model settings.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Each spawned process must check if xi , , xn, and , yn, can be derived from B and C, respectively.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Finally, the assignments of PoS tags to OOV segments is subject to language specific constraints relative to the token it was originated from.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Evalb is a Java re-implementation of the standard labeled precision\/recall metric.12 The ATB gives all punctuation a single tag.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Further, the special hash 0 suffices to flag empty buckets.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Each token may admit multiple analyses, each of which a sequence of one or more lexemes (we use li to denote a lexeme) belonging a presupposed Hebrew lexicon LEX.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Otherwise for the predecessor search hypothesis, we would have chosen a position that would not have been among the first n uncovered positions.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We use graph-based label propagation for cross-lingual knowledge transfer and use the projected labels as features in an unsupervised model (Berg- Kirkpatrick et al., 2010).","label":1},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"f1; ; Jg denotes a coverage set including all positions from the starting position 1 to position J and j 2 fJ \u00f4\u0080\u0080\u0080L; ; Jg.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Clearly this is not the only way to estimate word-frequencies, however, and one could consider applying other methods: in partic\u00c2\u00ad ular since the problem is similar to the problem of assigning part-of-speech tags to an untagged corpus given a lexicon and some initial estimate of the a priori probabilities for the tags, one might consider a more sophisticated approach such as that described in Kupiec (1992); one could also use methods that depend on a small hand-tagged seed corpus, as suggested by one reviewer.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"It is striking that from this point of view many formalisms can be grouped together as having identically structured derivation tree sets.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We have presented two general approaches to studying parser combination: parser switching and parse hybridization.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Previous approaches have tried to find examples that are similar to the target domain.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The approach builds from an initial seed set for a category, and is quite similar to the decision list approach described in (Yarowsky 95).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"To check for this, we do pairwise bootstrap resampling: Again, we repeatedly sample sets of sentences, this time from both systems, and compare their BLEU scores on these sets.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This has the potential drawback of increasing the number of features, which can make MERT less stable (Foster and Kuhn, 2009).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"That is, given a choice between segmenting a sequence abc into abc and ab, c, the former will always be picked so long as its cost does not exceed the summed costs of ab and c: while; it is possible for abc to be so costly as to preclude the larger grouping, this will certainly not usually be the case.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Since the work tapes store integers (which can be written in binary) that never exceed the size of the input, no configuration has space exceeding 0(log n).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(7) is at 0 when: 1) Vi : sign(gi (xi)) = sign(g2 (xi)); 2) Ig3(xi)l oo; and 3) sign(gi (xi)) = yi for i = 1, , m. In fact, Zco provides a bound on the sum of the classification error of the labeled examples and the number of disagreements between the two classifiers on the unlabeled examples.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The ratio of buckets to entries is controlled by space multiplier m > 1.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"As noted in Section 1, our code finds the longest matching entry wnf for query p(wn|s(wn\u22121 f ) The probability p(wn|wn\u22121 f ) is stored with wnf and the backoffs are immediately accessible in the provided state s(wn\u22121 When our code walks the data structure to find wnf , it visits wnn, wnn\u22121, ... , wnf .","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Label propagation is used to propagate these tags inwards and results in tag distributions for the middle word of each Italian trigram.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The only supervision is in the form of 7 seed rules (namely, that New York, California and U.S. are locations; that any name containing Mr is a person; that any name containing Incorporated is an organization; and that I.B.M. and Microsoft are organizations).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"During coreference resolution, BABAR checks (1) whether the anaphor is among the lexical expectations for the caseframe that extracts the candidate antecedent, and (2) whether the candidate is among the lexical expectations for the caseframe that extracts the anaphor.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Note that Zt is a normalization constant that ensures the distribution Dt+i sums to 1; it is a function of the weak hypothesis ht and the weight for that hypothesis at chosen at the tth round.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Since foreign names can be of any length, and since their original pronunciation is effectively unlimited, the identi\u00c2\u00ad fication of such names is tricky.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"If so, the CF Network reports that the anaphor and candidate may be coreferent.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Denote the unthresholded classifiers after t \u2014 1 rounds by git\u20141 and assume that it is the turn for the first classifier to be updated while the second one is kept fixed.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"It then computes a normalized Levenshtein edit distance between the extracted chain and the reference.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Almost all annotators expressed their preference to move to a ranking-based evaluation in the future.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Model Overview The model starts by generating a tag assignment T for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"For Experiment 1 it is meaningless as a baseline, since it would result in 0% accuracy. mation on path labels but drop the information about the syntactic head of the lifted arc, using the label d\u2191 instead of d\u2191h (AuxP\u2191 instead of AuxP\u2191Sb).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The only supervision is in the form of 7 seed rules (namely, that New York, California and U.S. are locations; that any name containing Mr is a person; that any name containing Incorporated is an organization; and that I.B.M. and Microsoft are organizations).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In section 5, we then evaluate the entire parsing system by training and evaluating on data from the Prague Dependency Treebank.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We focus on this difference between the tree sets of CFG's and IG's, and formalize the notion of dependence between paths in a tree set in Section 3.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Assuming unseen objects within each class are equiprobable, their probabilities are given by the Good-Turing theorem as: cis E( n'J.ls) Po oc N * E(N8ls) (2) where p815 is the probability of one unseen hanzi in class cls, E(n'J.15 ) is the expected number of hanzi in cls seen once, N is the total number of hanzi, and E(N(\/ 5 ) is the expected number of unseen hanzi in class cls.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, we have reason to doubt Chang et al.'s performance claims.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"A detailed description of the search procedure used is given in this patent.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The computing time, the number of search errors, and the multi-reference WER (mWER) are shown as a function of t0.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Table 2 shows the features used in the current version of the parser.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Thus, the arc from je to jedna will be labeled 5b\u2193 (to indicate that there is a syntactic head below it).","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"(Carlson, Marcu 2001) responded to this situation with relatively precise (and therefore long!)","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"This is the parse that is closest to the centroid of the observed parses under the similarity metric.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In the appositive case, the contextual predictor was the head of the modifying appositive (president in the Maury Cooper example); in the second case, the contextual predictor was the preposition together with the noun it modifies (plant_in in the Georgia example).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Any NLP application that presumes as input unrestricted text requires an initial phase of text analysis; such applications involve problems as diverse as machine translation, information retrieval, and text-to-speech synthesis (TIS).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Future work should also extend the approach to build a complete named entity extractor - a method that pulls proper names from text and then classifies them.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"A dynamic programming recursion similar to the one in Eq. 2 is evaluated.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We use a squared loss to penalize neighboring vertices that have different label distributions: kqi \u2212 qjk2 = Ey(qi(y) \u2212 qj(y))2, and additionally regularize the label distributions towards the uniform distribution U over all possible labels Y.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"This feature imposes constraints Table 1: Features based on the token string that are based on the probability of each name class during training.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Suppose M has to determine whether the k substrings ,.. .,ak can be derived from some symbol A.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We will describe the evaluation of such clusters in the next subsection.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Finally, Section 5 explains how BABAR relates to previous work, and Section 6 summarizes our conclusions.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Recent research has demonstrated that incorporating this sparsity constraint improves tagging accuracy.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We propose maximizing , where is the sequence of named- entity tags assigned to the words in the sentence , and is the information that can be extracted from the whole document containing . Our system is built on a maximum entropy classifier.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Sequence of Initial Caps (SOIC): In the sentence Even News Broadcasting Corp., noted for its accurate reporting, made the erroneous announcement., a NER may mistake Even News Broadcasting Corp. as an organization name.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"By using the output of a hand-coded system such as Proteus, MENE can improve its performance, and can even outperform IdentiFinder (Borthwick, 1999).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In this work, we take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":":zhong1 : 0.0 tjl :huo2 :0.0 (R:spub:\/ic of Ch:ina) + .,_,...I : jlong4 :0.0 (mUifaty genG181) 0 \u00c2\u00a3: _NC: 40.0 Figure 3 Partial Chinese Lexicon (NC = noun; NP = proper noun).c=- - I \u00e2\u0080\u00a2=- :il: .;ss:;zhangt \u00e2\u0080\u00a2 '-:.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The model described here thus demonstrates great potential for use in widespread applications.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"splitPUNC restores the convention of the WSJ.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"It uses a maximum entropy framework and classifies each word given its features.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Employing a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions is not only theoretically clean and linguistically justified and but also probabilistically apropriate and empirically sound.","label":1},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(student+plural) 'students,' which is derived by the affixation of the plural affix f, menD to the nounxue2shengl.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"In addition, this formulation results in a dramatic reduction in the number of model parameters thereby, enabling unusually rapid training.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We see from these results that the behavior of the parametric techniques are robust in the presence of a poor parser.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"It can also be seen clearly in this plot that two of the Taiwan speakers cluster very closely together, and the third Tai\u00c2\u00ad wan speaker is also close in the most significant dimension (the x axis).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This has led previous workers to adopt ad hoc linear weighting schemes (Finch and Sumita, 2008; Foster and Kuhn, 2007; L\u00a8u et al., 2007).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In this paper k = 3 (the three labels are person, organization, location), and we set a = 0.1.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The question is how to normalize the probabilities in such a way that smaller groupings have a better shot at winning.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This has the potential drawback of increasing the number of features, which can make MERT less stable (Foster and Kuhn, 2009).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We address the question of whether or not a formalism can generate only structural descriptions with independent paths.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.2 Contextual Role Knowledge.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"In addition, this formulation results in a dramatic reduction in the number of model parameters thereby, enabling unusually rapid training.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In (2a), we want to split the two morphemes since the correct analysis is that we have the adverb :1 cai2 'just,' the modal verb neng2 'be able' and the main verb R: Hke4fu2 'overcome'; the competing analysis is, of course, that we have the noun :1 cai2neng2 'talent,' followed by }'lijke4fu2 'overcome.'","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"For comparison to information-retrieval inspired baselines, eg (L\u00a8u et al., 2007), we select sentences from OUT using language model perplexities from IN.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"!!:\\ :yu2 e:_nc [::!!:zen3 l!f :moO t:_adv il!:shuot ,:_vb i i i 1 \u00e2\u0080\u00a2 10.03 13...","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Terrorism systems must distinguish between people who perpetrate a crime and people who are victims of a crime.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The feature-HMM model works better for all languages, generalizing the results achieved for English by Berg-Kirkpatrick et al. (2010).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"For example, syntactic decoders (Koehn et al., 2007; Dyer et al., 2010; Li et al., 2009) perform dynamic programming parametrized by both backward- and forward-looking state.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"IdentiFinder ' 99' s results are considerably better than IdentiFinder ' 97' s. IdentiFinder' s performance in MUC7 is published in (Miller et al., 1998).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"And if one is interested in TIS, one would probably consider the single orthographic word ACL to consist of three phonological words-lei s'i d\/-corresponding to the pronunciation of each of the letters in the acronym.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"It is a sequence of proper nouns within an NP; its last word Cooper is the head of the NP; and the NP has an appositive modifier (a vice president at S.&P.) whose head is a singular noun (president).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We evaluated BABAR on two domains: terrorism and natural disasters.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"from the subset of the United Informatics corpus not used in the training of the models.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"While Gan's system incorporates fairly sophisticated models of various linguistic information, it has the drawback that it has only been tested with a very small lexicon (a few hundred words) and on a very small test set (thirty sentences); there is therefore serious concern as to whether the methods that he discusses are scalable.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"7 Acknowledgements.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"7 www.cis.upenn.edu\/\u00e2\u0088\u00bcpdtb\/ 8 www.eml-research.de\/english\/Research\/NLP\/ Downloads had to buy a new car.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The average individual parser accuracy was reduced by more than 5% when we added this new parser, but the precision of the constituent voting technique was the only result that decreased significantly.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"where the husband's family name is optionally prepended to the woman's full name; thus ;f:*lf#i xu3lin2-yan2hai3 would represent the name that Ms. Lin Yanhai would take if she married someone named Xu.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Table 2 shows these similarity measures.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, there is a strong relationship between ni1s and the number of hanzi in the class.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This extends previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and using a simpler training procedure.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The use of the Good-Turing equation presumes suitable estimates of the unknown expectations it requires.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The following auxiliary quantity is defined: Qe0 (e; C; j) := probability of the best partial hypothesis (ei 1; bi 1), where C = fbkjk = 1; ; ig, bi = j, ei = e and ei\u00f4\u0080\u0080\u00801 = e0.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The development of the very first Hebrew Treebank (Sima\u2019an et al., 2001) called for the exploration of general statistical parsing methods, but the application was at first limited.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Our coreference resolver also incorporates an existential noun phrase recognizer and a DempsterShafer probabilistic model to make resolution decisions.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Although these authors report better gains than ours, they are with respect to a non-adapted baseline.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"A similar maximumlikelihood approach was used by Foster and Kuhn (2007), but for language models only.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A non-optimal analysis is shown with dotted lines in the bottom frame.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Similarly, the tokens and are tested against each list, and if found, a corresponding feature will be set to 1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"3.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Further, Maamouri and Bies (2004) argued that the English guidelines generalize well to other languages.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The accuracy of the sets in representing paraphrase ranged from 73% to 99%, depending on the NE categories and set sizes; the accuracy of the links for two evaluated domains was 73% and 86%.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This is a standard adaptation problem for SMT.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"(2009).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"While sorted arrays could be used to implement the same data structure as PROBING, effectively making m = 1, we abandoned this implementation because it is slower and larger than a trie implementation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"from the subset of the United Informatics corpus not used in the training of the models.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Sentence (2) and (3) help to disambiguate one way or the other.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Evaluation of the Segmentation as a Whole.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"3.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In this section, we will explain the algorithm step by step with examples.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"A Brief Introduction to the Chinese Writing System Most readers will undoubtedly be at least somewhat familiar with the nature of the Chinese writing system, but there are enough common misunderstandings that it is as well to spend a few paragraphs on properties of the Chinese script that will be relevant to topics discussed in this paper.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Word frequencies are estimated by a re-estimation procedure that involves apply\u00c2\u00ad ing the segmentation algorithm presented here to a corpus of 20 million words,8 using 8 Our training corpus was drawn from a larger corpus of mixed-genre text consisting mostly of.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"2.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Across all languages, high performance can be attained by selecting a single tag per word type.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"37.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In Input: {(x1,i, Initialize: Vi, j : e(xi) = 0.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"0 68.4 48.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(Riloff and Jones 99) was brought to our attention as we were preparing the final version of this paper.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Figure 1 shows sample sentences from these domains, which are widely divergent.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"However, it is also mostly political content (even if not focused on the internal workings of the European Union) and opinion.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"After adding a ROOT node to all trees, we train a grammar using six split-and- merge cycles and no Markovization.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Caching for IRSTLM is smaller at 0.09 GB resident memory, though it supports only a single thread.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Let s = a + b.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Xim, } associated with the ith example.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"In addition to the optimizations specific to each datastructure described in Section 2, we implement several general optimizations for language modeling.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We have described two data structures for language modeling that achieve substantial reductions in time and memory cost.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"7 68.3 56.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"4.1 The Task and the Corpus.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Each connected path (l1 ... lk) E L corresponds to one morphological segmentation possibility of W. The Parser Given a sequence of input tokens W = w1 ... wn and a morphological analyzer, we look for the most probable parse tree \u03c0 s.t.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Search hypotheses are processed separately according to their coverage set C. The best scored hypothesis for each coverage set is computed: QBeam(C) = max e;e0 ;S;j Qe0 (e; S; C; j) The hypothesis (e0; e; S; C; j) is pruned if: Qe0 (e; S; C; j) < t0 QBeam(C); where t0 is a threshold to control the number of surviving hypotheses.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For example, in the sentence that starts with \u00e2\u0080\u009cBush put a freeze on . . .","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Second, we show that although the Penn Arabic Treebank is similar to other tree- banks in gross statistical terms, annotation consistency remains problematic.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Recently, combination techniques have been investigated for part of speech tagging with positive results (van Halteren et al., 1998; Brill and Wu, 1998).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"A crucial difference is that the number of parameters is greatly reduced as is the number of variables that are sampled during each iteration.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"MENE (without the help of hand-coded systems) has been shown to be somewhat inferior in performance.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The following three sections elaborate these different stages is more detail.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Table 9 shows that MADA produces a high quality segmentation, and that the effect of cascading segmentation errors on parsing is only 1.92% F1.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Among machine learning-based NERs, Identi- Finder has proven to be the best on the official MUC6 and MUC7 test data.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The sequence of states needed to carry out the word reordering example in Fig.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Our \u201cProjection\u201d baseline is able to benefit from the bilingual information and greatly improves upon the monolingual baselines, but falls short of the \u201cNo LP\u201d model by 2.5% on an average.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The correct ambiguity resolution of the syntactic level therefore helps to resolve the morphological one, and vice versa.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"See Table 2 for the tag set size of other languages.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We smooth Prf(p \u2014* (s, p)) for rare and OOV segments (s E l, l E L, s unseen) using a \u201cper-tag\u201d probability distribution over rare segments which we estimate using relative frequency estimates for once-occurring segments.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We thank United Informatics for providing us with our corpus of Chinese text, and BDC for the 'Behavior ChineseEnglish Electronic Dictionary.'","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"As we will see from Table 3, not much improvement is derived from this feature.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We have developed a coreference resolver called BABAR that uses contextual role knowledge to make coreference decisions.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Gazdar (1985) considers a restriction of IG's in which no more than one nonterminal on the right-hand-side of a production can inherit the stack from the left-hand-side.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"However, there is again local grammatical information that should favor the split in the case of (1a): both .ma3 'horse' and .ma3 lu4 are nouns, but only .ma3 is consistent with the classifier pil, the classifier for horses.21 By a similar argument, the preference for not splitting , lm could be strengthened in (lb) by the observation that the classifier 1'1* tiao2 is consistent with long or winding objects like , lm ma3lu4 'road' but not with,ma3 'horse.'","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We call this approach parser switching.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Our results show that MENERGI performs as well as IdentiFinder when trained on comparable amount of training data.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"When the connective is an adverbial, there is much less clarity as to the range of the spans.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Hieu Hoang named the code \u201cKenLM\u201d and assisted with Moses along with Barry Haddow.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The most frequent NE category pairs are \u00e2\u0080\u009cPerson - Person (209,236), followed by \u00e2\u0080\u009cCountry - Coun- try\u00e2\u0080\u009d (95,123) and \u00e2\u0080\u009cPerson - Country\u00e2\u0080\u009d (75,509).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"While sorted arrays could be used to implement the same data structure as PROBING, effectively making m = 1, we abandoned this implementation because it is slower and larger than a trie implementation.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The general-language features have a slight advantage over the similarity features, and both are better than the SVM feature.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"1 53.8 47.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Section 4.1 explained that state s is stored by applications with partial hypotheses to determine when they can be recombined.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In practice, this sparsity constraint is difficult to incorporate in a traditional POS induction system (Me\u00c2\u00b4rialdo, 1994; Johnson, 2007; Gao and Johnson, 2008; Grac\u00c2\u00b8a et al., 2009; Berg-Kirkpatrick et al., 2010).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We can now compare this algorithm to that of (Yarowsky 95).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"As is standard, we use a fixed constant K for the number of tagging states.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Next, for each pair of NE categories, we collect all the contexts and find the keywords which are topical for that NE category pair.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We are also grateful to Markus Dickinson, Ali Farghaly, Nizar Habash, Seth Kulick, David McCloskey, Claude Reichard, Ryan Roth, and Reut Tsarfaty for constructive discussions.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"No: Predecessor coverage set Successor coverage set 1 (f1; ;mg n flg ; l0) !","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"), which precludes a single universal approach to adaptation.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We are especially grateful to Taylor Berg- Kirkpatrick for running additional experiments.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"A position is presented by the word at that position.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"For each language and setting, we report one-to-one (11) and many- to-one (m-1) accuracies.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This class-based model gives reasonable results: for six radical classes, Table 1 gives the estimated cost for an unseen hanzi in the class occurring as the second hanzi in a double GIVEN name.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"As with lexical expections, the semantic classes of co-referring expressions are 4 They may not be perfectly substitutable, for example one NP may be more specific (e.g., \u00e2\u0080\u009che\u00e2\u0080\u009d vs. \u00e2\u0080\u009cJohn F. Kennedy\u00e2\u0080\u009d).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We quantify error categories in both evaluation settings.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Because the Bikel parser has been parameter- ized for Arabic by the LDC, we do not change the default model settings.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The weak hypothesis can abstain from predicting the label of an instance x by setting h(x) = 0.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"We have used the Java-based opennlp maximum entropy package1.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Lexicon and OOV Handling Our data-driven morphological-analyzer proposes analyses for unknown tokens as described in Section 5.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The PROBING data structure is a rather straightforward application of these hash tables to store Ngram language models.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This token may further embed into a larger utterance, e.g., \u2018bcl hneim\u2019 (literally \u201cin-the-shadow the-pleasant\u201d, meaning roughly \u201cin the pleasant shadow\u201d) in which the dominated Noun is modified by a proceeding space-delimited adjective.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"It is difficult when IN and OUT are dissimilar, as they are in the cases we study.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Here, the pruning threshold t0 = 10:0 is used.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Extract NE instance pairs with contexts First, we extract NE pair instances with their context from the corpus.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"At most one feature in this group will be set to 1.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"This group contains a large number of features (one for each token string present in the training data).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"(Carlson, Marcu 2001) responded to this situation with relatively precise (and therefore long!)","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"wo rd => na m e 2.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"A different but supplementary perspective on discourse-based information structure is taken 11ventionalized patterns (e.g., order of informa by one of our partner projects, which is inter tion in news reports).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"One implementation issue deserves some elaboration.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"When the parser is trained on the transformed data, it will ideally learn not only to construct projective dependency structures but also to assign arc labels that encode information about lifts.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"A related point is that mutual information is helpful in augmenting existing electronic dictionaries, (cf.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(a) IDictionary D I D:d\/0.000 B:b\/0.000 B:b\/0.000 ( b ) ( c ) ( d ) I B e s t P a t h ( I d ( I ) o D * ) I cps:nd4.!l(l() Figure 2 An abstract example illustrating the segmentation algorithm.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The DL-CoTrain algorithm can be motivated as being a greedy method of satisfying the above 2 constraints.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Mai.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"This stage of label propagation results in a tag distribution ri over labels y, which encodes the proportion of times the middle word of ui E Vf aligns to English words vy tagged with label y: The second stage consists of running traditional label propagation to propagate labels from these peripheral vertices Vf\ufffd to all foreign language vertices in the graph, optimizing the following objective: 5 POS Induction After running label propagation (LP), we compute tag probabilities for foreign word types x by marginalizing the POS tag distributions of foreign trigrams ui = x\u2212 x x+ over the left and right context words: where the qi (i = 1, ... , |Vf|) are the label distributions over the foreign language vertices and \u00b5 and \u03bd are hyperparameters that we discuss in \u00a76.4.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"When a collision occurs, linear probing places the entry to be inserted in the next (higher index) empty bucket, wrapping around as necessary.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"10 Chinese speakers may object to this form, since the suffix f, menD (PL) is usually restricted to.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Automatic Paraphrase Discovery based on Context and Keywords between NE Pairs","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"A consequence of the ability to generate tree sets with this property is that CC's under this definition can generate the following language which can not be generated by either TAG's or HG's.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Their theoretical finding is simply stated: classification error rate decreases toward the noise rate exponentially in the number of independent, accurate classifiers.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The model starts by generating a tag assignment for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"While it is possible to derive a closed form solution for this convex objective function, it would require the inversion of a matrix of order |Vf|.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For the examples given in (1) and (2) this certainly seems possible.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(See also Wu and Fung [1994].)","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Finally, we provide a realistic eval uation in which segmentation is performed both in a pipeline and jointly with parsing (\u00c2\u00a76).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"can expect famous names like Zhou Enlai's to be in many dictionaries, but names such as :fi lf;f; shi2jil-lin2, the name of the second author of this paper, will not be found in any dictionary.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"3 60.7 50.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For each resolution in the training data, BABAR also associates the co-referring expression of an NP with the NP\u00e2\u0080\u0099s caseframe.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The form mnh itself can be read as at least three different verbs (\u201ccounted\u201d, \u201cappointed\u201d, \u201cwas appointed\u201d), a noun (\u201ca portion\u201d), and a possessed noun (\u201cher kind\u201d).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Finally, other packages implement language model estimation while we are currently dependent on them to generate an ARPA file.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"5 \u00e2\u0080\u0098Underspecified Rhetorical Markup Language\u00e2\u0080\u0099 6 This confirms the figure given by (Schauer, Hahn.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"2.2 Contextual Role Knowledge.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"However, it is unlikely that other occurrences of News Broadcasting Corp. in the same document also co-occur with Even.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2010)\u00e2\u0080\u0099s richest model: optimized via either EM or LBFGS, as their relative performance depends on the language.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"96 75.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We will briefly discuss this point in Section 3.1.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"This paper, however, provides a comprehensive overview of the data collection effort and its current state.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Using this heuristic, BABAR identifies existential definite NPs in the training corpus using our previous learning algorithm (Bean and Riloff, 1999) and resolves all occurrences of the same existential NP with each another.1 2.1.2 Syntactic Seeding BABAR also uses syntactic heuristics to identify anaphors and antecedents that can be easily resolved.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The dependency graph in Figure 1 satisfies all the defining conditions above, but it fails to satisfy the condition ofprojectivity (Kahane et al., 1998): The arc connecting the head jedna (one) to the dependent Z (out-of) spans the token je (is), which is not dominated by jedna.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Each unlabeled pair (x1,i, x2,i) is represented as an edge between nodes corresponding to x1,i and X2,i in the graph.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For each co-occurrence relation (noun\/caseframe for CFLex, and caseframe\/caseframe for CFNet), BABAR computes its log-likelihood value and looks it up in the \u00cf\u00872 table to obtain a confidence level.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The authors acknowledge the support of the NSF (CAREER grant IIS0448168, and grant IIS 0904684).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Reflexive pronouns with only 1 NP in scope..","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We experimented with increasingly rich grammars read off of the treebank.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In contrast to the Bayesian HMM, \u00ce\u00b8t is not drawn from a distribution which has support for each of the n word types.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"logical rules, and personal names; the transitive closure of the resulting machine is then computed.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The price shocked her.), or same-kind (e.g., Her health insurance paid for the hospital fees, but the automobile insurance did not cover the repair.).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"All annotations are done with specific tools and in XML; each layer has its own DTD.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"While the proportion of sentences containing non-projective dependencies is often 15\u201325%, the total proportion of non-projective arcs is normally only 1\u20132%.","label":0},{"source_text":"There is no global pruning.","target_text":"This work has been supported as part of the Verbmobil project (contract number 01 IV 601 A) by the German Federal Ministry of Education, Science, Research and Technology and as part of the Eutrans project (ESPRIT project number 30268) by the European Community.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Church and Hanks [1989]), and we have used lists of character pairs ranked by mutual information to expand our own dictionary.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"with the number of exactly matching guess trees.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(7), such as the likelihood function used in maximum-entropy problems and other generalized additive models (Lafferty 99).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"These are written to the state s(wn1) and returned so that they can be used for the following query.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"For the translation model Pr(fJ 1 jeI 1), we go on the assumption that each source word is aligned to exactly one target word.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"4 Traditional Arabic linguistic theory treats both of these types as subcategories of noun \u00ef\u00bf\u00bd '.i . Figure 1: The Stanford parser (Klein and Manning, 2002) is unable to recover the verbal reading of the unvocalized surface form 0 an (Table 1).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The number of latent HMM states for each language in our experiments was set to the number of fine tags in the language\u2019s treebank.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"html 4 www.wagsoft.com\/RSTTool assigning rhetorical relations is a process loaded with ambiguity and, possibly, subjectivity.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Three state-of-the-art statistical parsers are combined to produce more accurate parses, as well as new bounds on achievable Treebank parsing accuracy.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The F- measure score increased for both domains, reflecting a substantial increase in recall with a small decrease in precision.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"If is unique, then a feature (Unique, Zone) is set to 1, where Zone is the document zone where appears.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This causes a problem for reverse trie implementations, including SRILM itself, because it leaves n+1-grams without an n-gram node pointing to them.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The set is then compared with the set generated from the Penn Treebank parse to determine the precision and recall.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The overall performance of the LTG system was outstanding, but the system consists of a sequence of many hand-coded rules and machine-learning modules.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Although the best published results for the Collins parser is 80% UAS (Collins, 1999), this parser reaches 82% when trained on the entire training data set, and an adapted version of Charniak\u2019s parser (Charniak, 2000) performs at 84% (Jan Haji\u02c7c, pers. comm.).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The first unsupervised algorithm we describe is based on the decision list method from (Yarowsky 95).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each caseframe, BABAR collects the semantic classes associated with the head nouns of NPs that were extracted by the caseframe.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In this case, this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Even with vocalization, there are linguistic categories that are difficult to identify without semantic clues.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"shows some keywords with their scores.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"First, we will describe their method and compare it with our method.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"An additional case of super-segmental morphology is the case of Pronominal Clitics.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Obviously, the presence of a title after a potential name N increases the probability that N is in fact a name.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"3 Throughout this paper we shall give Chinese examples in traditional orthography, followed.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"36.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This paper describes the several performance techniques used and presents benchmarks against alternative implementations.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Figure 2 shows timing results.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The first shows how constituent features and context do not help in deciding which parser to trust.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"VBD she added VP PUNC \u00e2\u0080\u009c SBAR IN NP 0 NN.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"2 62.2 39.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The code is opensource, has minimal dependencies, and offers both C++ and Java interfaces for integration.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Je voudrais pr\u00b4eciser, a` l\u2019adresse du commissaire Liikanen, qu\u2019il n\u2019est pas ais\u00b4e de recourir aux tribunaux nationaux.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"3 54.4 33.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"All the links in the \u00e2\u0080\u009cCC-domain are shown in Step 4 in subsection 3.2.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Unfortunately, modifying the model to account for these kind of dependencies is not at all straightforward.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"(S; C; j); Not only the coverage set C and the positions j; j0, but also the verbgroup states S; S0 are taken into account.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Although we have not done any experiments on other languages, this way of using global features from a whole document should be applicable to other languages.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Instead of offsetting new topics with punctuation, writers of MSA in sert connectives such as \u00ef\u00bf\u00bd wa and \u00ef\u00bf\u00bd fa to link new elements to both preceding clauses and the text as a whole.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The linear LM (lin lm), TM (lin tm) and MAP TM (map tm) used with non-adapted counterparts perform in all cases slightly worse than the log-linear combination, which adapts both LM and TM components.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"37 79.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"(we ignored the 419 trees in their development set.)","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The core of Yarowsky's algorithm is as follows: where h is defined by the formula in equation 2, with counts restricted to training data examples that have been labeled in step 2.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We will return to these issues in the discussion section.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"As lower frequency examples include noise, we set a threshold that an NE category pair should appear at least 5 times to be considered and an NE instance pair should appear at least twice to be considered.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"For all these annotation taks, G\u00c2\u00a8otze developed a series of questions (essentially a decision tree) designed to lead the annotator to the ap propriate judgement.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For each pair of judges, consider one judge as the standard,.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"By establishing significantly higher parsing baselines, we have shown that Arabic parsing performance is not as poor as previously thought, but remains much lower than English.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Caching for IRSTLM is smaller at 0.09 GB resident memory, though it supports only a single thread.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Each vertex within a connected component must have the same label \u2014 in the binary classification case, we need a single labeled example to identify which component should get which label.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"It is difficult when IN and OUT are dissimilar, as they are in the cases we study.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Our test machine has two Intel Xeon E5410 processors totaling eight cores, 32 GB RAM, and four Seagate Barracuda disks in software RAID 0 running Linux 2.6.18.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The resulting parsers surpass the best previously published performance results for the Penn Treebank.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"This heuristics is used to prune all segmentation possibilities involving \u201clexically improper\u201d segments.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"By contrast, when we turn to a comparison of the three encoding schemes it is hard to find any significant differences, and the overall impression is that it makes little or no difference which encoding scheme is used, as long as there is some indication of which words are assigned their linear head instead of their syntactic head by the projective parser.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Initially, the DempsterShafer model assumes that all hypotheses are equally likely, so it creates a set called \u00ce\u00b8 that includes all hypotheses.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"These two properties of the tree sets are not only linguistically relevant, but also have computational importance.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Previous work on morphological and syntactic disambiguation in Hebrew used different sets of data, different splits, differing annotation schemes, and different evaluation measures.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"LCFRS's share several properties possessed by the class of mildly context-sensitive formalisms discussed by Joshi (1983\/85).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Following Sproat and Shih (1990), performance for Chinese segmentation systems is generally reported in terms of the dual measures of precision and recalP It is fairly standard to report precision and recall scores in the mid to high 90% range.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A straightforward way to find the shortest tour is by trying all possible permutations of the n cities.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Also surprising is the low test set OOV rate given the possibility of morphological variation in Arabic.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"3.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"immediately by a Romanization into the pinyin transliteration scheme; numerals following each pinyin syllable represent tones.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"2.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"First, we will describe their method and compare it with our method.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We concentrate on those sets.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The 13 positions of the source sentence are processed in the order shown.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The commentaries in PCC are all of roughly the same length, ranging from 8 to 10 sentences.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"2.2.2 The Caseframe Network The first type of contextual role knowledge that BABAR learns is the Caseframe Network (CFNet), which identifies caseframes that co-occur in anaphor\/antecedent resolutions.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"An analysis of nouns that occur in both the singular and the plural in our database reveals that there is indeed a slight but significant positive correlation-R2 = 0.20, p < 0.005; see Figure 6.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Applications The discovered paraphrases have multiple applications.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Across all languages, +PRIOR consistently outperforms 1TW, reducing error on average by 9.1% and 5.9% on best and median settings respectively.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The second case involves existential noun phrases (Allen, 1995), which are noun phrases that uniquely specify an object or concept and therefore do not need a prior referent in the discourse.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We thank members of the MIT NLP group for their suggestions and comments.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Table 2 shows our complete set of results.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In this situation, BABAR takes the conservative approach and declines to make a resolution.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This is similar to using the Linux MAP POPULATE flag that is our default loading mechanism.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For \u00c3\u0086 = 1, a new target language word is generated using the trigram language model p(eje0; e00).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We refer to (T , W ) as the lexicon of a language and \u00cf\u0088 for the parameters for their generation; \u00cf\u0088 depends on a single hyperparameter \u00ce\u00b2.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"In Table 1 we see with very few exceptions that the isolated constituent precision is less than 0.5 when we use the constituent label as a feature.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"So the success of the algorithm may well be due to its success in maximizing the number of unlabeled examples on which the two decision lists agree.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"There is a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099 of ten commentaries, for which the range of information (except for syntax) has been completed; the remaining data has been annotated to different degrees, as explained below.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"A multicomponent Tree Adjoining Grammar (MCTAG) consists of a finite set of finite elementary tree sets.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The search starts in hypothesis (f;g; 0) and ends in the hypotheses (f1; ; Jg; j), with j 2 f1; ; Jg.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We trained this model by optimizing the following objective function: Note that this involves marginalizing out all possible state configurations z for a sentence x, resulting in a non-convex objective.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Some approaches depend upon some form of con\u00c2\u00ad straint satisfaction based on syntactic or semantic features (e.g., Yeh and Lee [1991], which uses a unification-based approach).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In this section, we describe how contextual role knowledge is represented and learned.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The precision and recall of similarity switching and constituent voting are both significantly better than the best individual parser, and constituent voting is significantly better than parser switching in precision.4 Constituent voting gives the highest accuracy for parsing the Penn Treebank reported to date.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For example, syntactic decoders (Koehn et al., 2007; Dyer et al., 2010; Li et al., 2009) perform dynamic programming parametrized by both backward- and forward-looking state.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This is most severe with RandLM in the multi-threaded case, where each thread keeps a separate cache, exceeding the original model size.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We report the F1 value of both measures.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"This process is repeated 5 times by rotating the data appropriately.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Still, from a theoretical point of view, projective parsing of non-projective structures has the drawback that it rules out perfect accuracy even as an asymptotic goal.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"A summary of the corpus used in the experiments is given in Table 3.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Its success depends on the two domains being relatively close, and on the OUT corpus not being so large as to overwhelm the contribution of IN.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Clearly, for judges h and h taking h as standard and computing the precision and recall for Jz yields the same results as taking h as the standard, and computing for h, 14 All evaluation materials, with the exception of those used for evaluating personal names were drawn.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The state-to-tag mapping is obtained from the best hyperparameter setting for 11 mapping shown in Table 3.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Full Chinese personal names are in one respect simple: they are always of the form family+given.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For eight judges, ranging k between 1 and 8 corresponded to a precision score range of 90% to 30%, meaning that there were relatively few words (30% of those found by the automatic segmenter) on which all judges agreed, whereas most of the words found by the segmenter were such that one human judge agreed.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Future extensions of the system might include: 1) An extended translation model, where we use more context to predict a source word.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"F-measure is the harmonic mean of precision and recall, 2PR\/(P + R).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The segmenter handles the grouping of hanzi into words and outputs word pronunciations, with default pronunciations for hanzi it cannot group; we focus here primarily on the system's ability to segment text appropriately (rather than on its pronunciation abilities).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We therefore used the arithmetic mean of each interjudge precision-recall pair as a single measure of interjudge similarity.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"handled given appropriate models.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"In graph-based learning approaches one constructs a graph whose vertices are labeled and unlabeled examples, and whose weighted edges encode the degree to which the examples they link have the same label (Zhu et al., 2003).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"of Tokens MENERGI 318 160,000 200 180,000 IdentiFinder \u00e2\u0080\u0093 650,000 \u00e2\u0080\u0093 790,000 MENE \u00e2\u0080\u0093 \u00e2\u0080\u0093 350 321,000 Table 4: Training Data MUC7 test accuracy.2 For MUC6, the reduction in error due to global features is 27%, and for MUC7,14%.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"To set \u03b2, we used the same criterion as for \u03b1, over a dev corpus: The MAP combination was used for TM probabilities only, in part due to a technical difficulty in formulating coherent counts when using standard LM smoothing techniques (Kneser and Ney, 1995).3 Motivated by information retrieval, a number of approaches choose \u201crelevant\u201d sentence pairs from OUT by matching individual source sentences from IN (Hildebrand et al., 2005; L\u00a8u et al., 2007), or individual target hypotheses (Zhao et al., 2004).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Oracle results).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This is not unreasonable given the application to phrase pairs from OUT, but it suggests that an interesting alternative might be to use a plain log-linear weighting function exp(Ei Aifi(s, t)), with outputs in [0, oo].","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"However, it is also mostly political content (even if not focused on the internal workings of the European Union) and opinion.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Since the transducers are built from human-readable descriptions using a lexical toolkit (Sproat 1995), the system is easily maintained and extended.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The techniques we develop can be extended in a relatively straightforward manner to the more general case when OUT consists of multiple sub-domains.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Decreasing the threshold results in higher mWER due to additional search errors.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"A similar maximumlikelihood approach was used by Foster and Kuhn (2007), but for language models only.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"When a company buys another company, a paying event can occur, but these two phrases do not indicate the same event.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We developed a first version of annotation guidelines for co-reference in PCC (Gross 2003), which served as basis for annotating the core corpus but have not been empirically evaluated for inter-annotator agreement yet.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"For Experiment 1 it is meaningless as a baseline, since it would result in 0% accuracy. mation on path labels but drop the information about the syntactic head of the lifted arc, using the label d\u2191 instead of d\u2191h (AuxP\u2191 instead of AuxP\u2191Sb).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(a) IDictionary D I D:d\/0.000 B:b\/0.000 B:b\/0.000 ( b ) ( c ) ( d ) I B e s t P a t h ( I d ( I ) o D * ) I cps:nd4.!l(l() Figure 2 An abstract example illustrating the segmentation algorithm.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"As a (crude) approximation, we normalize the extraction patterns with respect to active and passive voice and label those extractions as agents or patients.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The portions of information in the large window can be individually clicked visible or invisible; here we have chosen to see (from top to bottom) \u00e2\u0080\u00a2 the full text, \u00e2\u0080\u00a2 the annotation values for the activated annotation set (co-reference), \u00e2\u0080\u00a2 the actual annotation tiers, and \u00e2\u0080\u00a2 the portion of text currently \u00e2\u0080\u0098in focus\u00e2\u0080\u0099 (which also appears underlined in the full text).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"For alif with hamza, normalization can be seen as another level of devocalization.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The difference in performance between pronouns and definite noun phrases surprised us.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This is a standard adaptation problem for SMT.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"This might be because our features are more comprehensive than those used by Borthwick.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The toplevel weights are trained to maximize a metric such as BLEU on a small development set of approximately 1000 sentence pairs.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"na me =>2 ha nzi fa mi ly 2 ha nzi gi ve n 5.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"(2006).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"It is a relatively frequent word in the domain, but it can be used in different extraction scenarios.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"87 Table 7: Test set results.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The equation for sampling a single type-level assignment Ti is given by, 0.2 0 5 10 15 20 25 30 Iteration Figure 2: Graph of the one-to-one accuracy of our full model (+FEATS) under the best hyperparameter setting by iteration (see Section 5).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Additionally, it works about 3 times as fast as the IBM style search.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Also, the argument has been made that machine translation performance should be evaluated via task-based evaluation metrics, i.e. how much it assists performing a useful task, such as supporting human translators or aiding the analysis of texts.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In this section, we will explain the algorithm step by step with examples.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"A possible probabilistic model for assigning probabilities to complex analyses of a surface form may be and indeed recent sequential disambiguation models for Hebrew (Adler and Elhadad, 2006) and Arabic (Smith et al., 2005) present similar models.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"In fact, training on the official training data is not suitable as the articles in this data set are entirely about aviation disasters, and the test data is about air vehicle launching.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The semantic caseframe expectations are used in two ways.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Either save money at any cost - or give priority to education.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The only way to handle such phenomena within the framework described here is simply to expand out the reduplicated forms beforehand, and incorporate the expanded forms into the lexical transducer.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Evaluation metrics used are Attachment Score (AS), i.e. the proportion of tokens that are attached to the correct head, and Exact Match (EM), i.e. the proportion of sentences for which the dependency graph exactly matches the gold standard.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The discovered paraphrases can be a big help to reduce human labor and create a more comprehensive pattern set.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Time starts when Moses is launched and therefore includes model loading time.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"4 To be sure, it is not always true that a hanzi represents a syllable or that it represents a morpheme.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"7 Acknowledgements.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Almost all annotators expressed their preference to move to a ranking-based evaluation in the future.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"These models generally outperform our memory consumption but are much slower, even when cached.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In that work, mutual information was used to decide whether to group adjacent hanzi into two-hanzi words.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Little attention, however, has been paid to the structural descriptions that these formalisms can assign to strings, i.e. their strong generative capacity.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Search hypotheses are processed separately according to their coverage set C. The best scored hypothesis for each coverage set is computed: QBeam(C) = max e;e0 ;S;j Qe0 (e; S; C; j) The hypothesis (e0; e; S; C; j) is pruned if: Qe0 (e; S; C; j) < t0 QBeam(C); where t0 is a threshold to control the number of surviving hypotheses.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We formulate the update as follows: where \u2200ui \u2208 Vf \\ Vfl, \u03b3i(y) and \u03bai are defined as: We ran this procedure for 10 iterations.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"This drastic tree manipulation is not appropriate for situations in which we want to assign particular structures to sentences.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In this way, the method reported on here will necessarily be similar to a greedy method, though of course not identical.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"For our experiments we also report the mean of precision and recall, which we denote by (P + R)I2 and F-measure.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The first two rows of the table are baselines.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The 2(Yarowsky 95) describes the use of more sophisticated smoothing methods.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Several methods exist to handle collisions; we use linear probing because it has less memory overhead when entries are small.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"4.3 Morphological Analysis.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Without using the same test corpus, direct comparison is obviously difficult; fortunately, Chang et al. include a list of about 60 sentence fragments that exemplify various categories of performance for their system.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As a first step towards modeling transliterated names, we have collected all hanzi occurring more than once in the roughly 750 foreign names in our dictionary, and we estimate the probabil\u00c2\u00ad ity of occurrence of each hanzi in a transliteration (pTN(hanzi;)) using the maximum likelihood estimate.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Presence of the determiner J Al. 2.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We measured recall (Rec), precision (Pr), and the F-measure (F) with recall and precision equally weighted.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Source sentence words are aligned with hypothesized target sentence words, where the choice of a new source word, which has not been aligned with a target word yet, is restricted1.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"8 1 2.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"96 75.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The search starts in hypothesis (f;g; 0) and ends in the hypotheses (f1; ; Jg; j), with j 2 f1; ; Jg.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The implication of this ambiguity for a parser is that the yield of syntactic trees no longer consists of spacedelimited tokens, and the expected number of leaves in the syntactic analysis in not known in advance.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"So, we set a threshold that at least two examples are required to build a link.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Step 2.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Interpolation search is therefore a form of binary search with better estimates informed by the uniform key distribution.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"While our method also enforces a singe tag per word constraint, it leverages the transition distribution encoded in an HMM, thereby benefiting from a richer representation of context.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"In the initial release of the ATB, inter-annotator agreement was inferior to other LDC treebanks (Maamouri et al., 2008).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"But we will show that the use of unlabeled data can drastically reduce the need for supervision.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"IRSTLM\u2019s quantized variant is the inspiration for our quantized variant.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Evaluation within a set The evaluation of paraphrases within a set of phrases which share a keyword is illustrated in Figure 4.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Note, however, that there might be situations in which Zco in fact increases.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The resulting structural differences between tree- banks can account for relative differences in parsing performance.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Figure 2 shows a screenshot (which is of somewhat limited value, though, as color plays a major role in signalling the different statuses of the information).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"3 Techniques for automatic vocalization have been studied (Zitouni et al., 2006; Habash and Rambow, 2007).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For each pair of judges consider one judge as the standard,.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"52 15.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"This group consists of (3 total number of possible zones) features.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"These three parsers have given the best reported parsing results on the Penn Treebank Wall Street Journal corpus (Marcus et al., 1993).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"3 68.9 50.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We tagged each noun with the top-level semantic classes assigned to it in Word- Net.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"11 www.ling.unipotsdam.de\/sfb\/projekt a3.php 12 This step was carried out in the course of the diploma thesis work of David Reitter (2003), which de serves special mention here.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Performance improvements transfer to the Moses (Koehn et al., 2007), cdec (Dyer et al., 2010), and Joshua (Li et al., 2009) translation systems where our code has been integrated.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The CFLex and CFNet knowledge sources provide positive evidence that a candidate NP and anaphor might be coreferent.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In the following, we use the notation wi wj to mean that (wi, r, wj) E A; r we also use wi wj to denote an arc with unspecified label and wi \u2014*\u2217 wj for the reflexive and transitive closure of the (unlabeled) arc relation.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Tokens tagged as PUNC are not discarded unless they consist entirely of punctuation.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Given a key k, it estimates the position If the estimate is exact (A[pivot] = k), then the algorithm terminates succesfully.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In the experiments below, we employ a data-driven deterministic dependency parser producing labeled projective dependency graphs,3 previously tested on Swedish (Nivre et al., 2004) and English (Nivre and Scholz, 2004).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Their default variant implements a forward trie, in which words are looked up in their natural left-to-right order.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In the appositive case, the contextual predictor was the head of the modifying appositive (president in the Maury Cooper example); in the second case, the contextual predictor was the preposition together with the noun it modifies (plant_in in the Georgia example).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Maamouri et al.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The maximum precision oracle is an upper bound on the possible gain we can achieve by parse hybridization.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Having explained the various layers of annotation in PCC, we now turn to the question what all this might be good for.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Consider the following sentences: (a) Jose Maria Martinez, Roberto Lisandy, and Dino Rossy, who were staying at a Tecun Uman hotel, were kidnapped by armed men who took them to an unknown place.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Combining multiple highly-accurate independent parsers yields promising results.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We did have a problem classifying some modified noun phrases where the modified phrase does not represent a qualified or restricted form of the head, like \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, as these are both represented by the keyword \u00e2\u0080\u009cchairman\u00e2\u0080\u009d.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Ablation Analysis We evaluate the impact of incorporating various linguistic features into our model in Table 3.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The same argument holds for resolving PP attachment of a prefixed preposition or marking conjunction of elements of any kind.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"There are two key benefits of this model architecture.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We use the log-linear tagger of Toutanova et al.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"A list of words occurring more than 10 times in the training data is also collected (commonWords).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, there are several reasons why this approach will not in general work: 1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In the numerator, however, the counts of ni1s are quite irregular, in\u00c2\u00ad cluding several zeros (e.g., RAT, none of whose members were seen).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"gaolxing4 'happy'=> F.i'JF.i'JJI!JI!","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We elected run Moses single-threaded to minimize the impact of RandLM\u2019s cache on memory use.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"A tree set may be said to have dependencies between paths if some &quot;appropriate&quot; subset can be shown to have dependent paths as defined above.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"They first collect the NE instance pairs and contexts, just like our method.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The Danish Dependency Treebank (DDT) comprises about 100K words of text selected from the Danish PAROLE corpus, with annotation of primary and secondary dependencies (Kromann, 2003).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The first point we need to address is what type of linguistic object a hanzi repre\u00c2\u00ad sents.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"However, since we want to preserve as much of the original structure as possible, we are interested in finding a transformation that involves a minimal number of lifts.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Here, the term frequency (TF) is the frequency of a word in the bag and the inverse term frequency (ITF) is the inverse of the log of the frequency in the entire corpus.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"With the exception of the Dutch data set, no other processing is performed on the annotated tags.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"While the proportion of sentences containing non-projective dependencies is often 15\u201325%, the total proportion of non-projective arcs is normally only 1\u20132%.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Instead, the designs of the various annotation layers and the actual annotation work are results of a series of diploma theses, of students\u00e2\u0080\u0099 work in course projects, and to some extent of paid assistentships.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Equ.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"3 Throughout this paper we shall give Chinese examples in traditional orthography, followed.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Lexicalizing several POS tags improves performance.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Sc opi ng filters candidate if outside the anaphor\u00e2\u0080\u0099s scope.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Then, the German infinitive 'besuchen' and the negation particle 'nicht' are translated.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"In addition to using the simple features directly, we also trained an SVM classifier with these features to distinguish between IN and OUT phrase pairs.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(Riloff and Shepherd 97) describe a bootstrapping approach for acquiring nouns in particular categories (such as &quot;vehicle&quot; or &quot;weapon&quot; categories).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Fortunately, performance was stable across various values, and we were able to use the same hyperparameters for all languages.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"0 D\/ nc 5.0 The minimal dictionary encoding this information is represented by the WFST in Figure 2(a).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Two measures that can be used to compare judgments are: 1.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":" We evaluate the system's performance by comparing its segmentation 'Tudgments\" with the judgments of a pool of human segmenters, and the system is shown to perform quite well.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"A spelling rule might be a simple look-up for the string (e.g., a rule that Honduras is a location) or a rule that looks at words within a string (e.g., a rule that any string containing Mr. is a person).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"A summary of the corpus used in the experiments is given in Table 3.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"BABAR employs information extraction techniques to represent and learn role relationships.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Methods for expanding the dictionary include, of course, morphological rules, rules for segmenting personal names, as well as numeral sequences, expressions for dates, and so forth (Chen and Liu 1992; Wang, Li, and Chang 1992; Chang and Chen 1993; Nie, Jin, and Hannan 1994).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Given parameter estimates, the label for a test example x is defined as We should note that the model in equation 9 is deficient, in that it assigns greater than zero probability to some feature combinations that are impossible.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"First, we will describe their method and compare it with our method.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"For developers of Statistical Machine Translation (SMT) systems, an additional complication is the heterogeneous nature of SMT components (word-alignment model, language model, translation model, etc.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Even if an example like this is not labeled, it can be interpreted as a &quot;hint&quot; that Mr and president imply the same category.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We extend Subramanya et al.\u2019s intuitions to our bilingual setup.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Our TRIE implementation is designed to improve upon IRSTLM using a reverse trie with improved search, bit level packing, and stateful queries.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"There any many techniques for improving language model speed and reducing memory consumption.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Then, it can be verified that We can now derive the CoBoost algorithm as a means of minimizing Zco.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Location list is processed into a list of unigrams and bigrams (e.g., New York).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For example, in .., says Mr. Cooper, a vice president of.. both a spelling feature (that the string contains Mr.) and a contextual feature (that president modifies the string) are strong indications that Mr. Cooper is of type Person.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This withdrawal by the treasury secretary is understandable, though.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"0 70.9 42.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"9 61.0 44.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In this section, we try to compare our results with those obtained by IdentiFinder ' 97 (Bikel et al., 1997), IdentiFinder ' 99 (Bikel et al., 1999), and MENE (Borthwick, 1999).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"T e r r o r i s m Ca sef ra me Semantic Classes <a ge nt > ass ass ina ted group, human inv esti gat ion int o < N P> event exp lod ed out sid e < N P> building N a t u r a l D i s a s t e r s Ca sef ra me Semantic Classes <a ge nt > inv esti gat ing cau se group, human sur viv or of < N P> event, natphenom hit wit h < N P> attribute, natphenom Figure 3: Semantic Caseframe Expectations Figure 2: Lexical Caseframe Expectations To illustrate how lexical expectations are used, suppose we want to determine whether noun phrase X is the antecedent for noun phrase Y. If they are coreferent, then X and Y should be substitutable for one another in the story.4 Consider these sentences: (S1) Fred was killed by a masked man with a revolver.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"See Table 2 for the tag set size of other languages.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Unlabeled examples in the named-entity classification problem can reduce the need for supervision to a handful of seed rules.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"A similar structure is used in speech recognition.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Contextual role knowledge provides evidence as to whether a candidate is a plausible antecedent for an anaphor.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This is because our corpus is not annotated, and hence does not distinguish between the various words represented by homographs, such as, which could be \/adv jiangl 'be about to' orInc jiang4 '(military) general'-as in 1j\\xiao3jiang4 'little general.'","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"For example, one of the ATB samples was the determiner -\"\" ; dhalik\u00e2\u0080\u009cthat.\u00e2\u0080\u009d The sample occurred in 1507 corpus po sitions, and we found that the annotations were consistent.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"One class comprises words derived by productive morphologi\u00c2\u00ad cal processes, such as plural noun formation using the suffix ir, menD.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"MENE (without the help of hand-coded systems) has been shown to be somewhat inferior in performance.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"\u00e2\u0080\u0094ested in correlations between prosody and dis course structure.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Our evaluation includes both weighted and un- weighted lattices.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In principle a feature could be an arbitrary predicate of the (spelling, context) pair; for reasons that will become clear, features are limited to querying either the spelling or context alone.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If is unique, then a feature (Unique, Zone) is set to 1, where Zone is the document zone where appears.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Because all English vertices are going to be labeled, we do not need to disambiguate them by embedding them in trigrams.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"However, it is desirable if we can separate them.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Mohri [1995]) shows promise for improving this situation.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Before describing the unsupervised case we first describe the supervised version of the algorithm: Input to the learning algorithm: n labeled examples of the form (xi, y\u201e). y, is the label of the ith example (given that there are k possible labels, y, is a member of y = {1 ... 0). xi is a set of mi features {x,1, Xi2 .","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Methods that allow multiple segmentations must provide criteria for choosing the best segmentation.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The Expectation Maximization (EM) algorithm (Dempster, Laird and Rubin 77) is a common approach for unsupervised training; in this section we describe its application to the named entity problem.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In the maximum entropy framework, there is no such constraint.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"It is the performance we could achieve if an omniscient observer told us which parser to pick for each of the sentences.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The first experiment uses data from two dependency treebanks.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"One of the difficulties in Natural Language Processing is the fact that there are many ways to express the same thing or event.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"This procedure is repeated for T rounds while alternating between the two classifiers.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"For the TM, this is: where cI(s, t) is the count in the IN phrase table of pair (s, t), po(s|t) is its probability under the OUT TM, and cI(t) = &quot;s, cI(s', t).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(8) can now be rewritten5 as which is of the same form as the function Zt used in AdaBoost.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Such sequences are given additional features of A begin, A continue, or A end, and the acronym is given a feature A unique.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In the rhetorical tree, nuclearity information is then used to extract a \u00e2\u0080\u009ckernel tree\u00e2\u0080\u009d that supposedly represents the key information from which the summary can be generated (which in turn may involve co-reference information, as we want to avoid dangling pronouns in a summary).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Raj and Whittaker (2003) show that integers in a trie implementation can be compressed substantially.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Again we notice that the isolated constituent precision is larger than 0.5 only in those partitions that contain very few samples.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Given a sufficient number of randomly drawn unlabeled examples (i.e., edges), we will induce two completely connected components that together span the entire graph.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This is especially bad with PROBING because it is based on hashing and performs random lookups, but it is not intended to be used in low-memory scenarios.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"[Hasegawa et al. 04] reported only on relation discovery, but one could easily acquire para phrases from the results.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"More examples are shown in Figure 5.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"However, there will remain a large number of words that are not readily adduced to any produc\u00c2\u00ad tive pattern and that would simply have to be added to the dictionary.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Reiche\u00e2\u0080\u0099s colleagues will make sure that the concept is waterproof.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR uses two methods to identify anaphors that can be easily and reliably resolved with their antecedent: lexical seeding and syntactic seeding.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"When this metric is less than 0.5, we expect to incur more errors' than we will remove by adding those constituents to the parse.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Although this feature helps, we encounter one consequence of variable word order.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"One way to approach this discrepancy is to assume a preceding phase of morphological segmentation for extracting the different lexical items that exist at the token level (as is done, to the best of our knowledge, in all parsing related work on Arabic and its dialects (Chiang et al., 2006)).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Then each arc of D maps either from an element of H to an element of p, or from E-i.e., the empty string-to an element of P. More specifically, each word is represented in the dictionary as a sequence of arcs, starting from the initial state of D and labeled with an element 5 of Hxp, which is terminated with a weighted arc labeled with an element of Ex P. The weight represents the estimated cost (negative log probability) of the word.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The state in future has not enough work for its many teachers.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"com t 600 Mountain Avenue, 2c278, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"But we follow the more direct adaptation of Evalb suggested by Tsarfaty (2006), who viewed exact segmentation as the ultimate goal.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"I \u00e2\u0080\u00a2 JAPANS :rl4 .\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\)\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"o\u00c2\u00b7'\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\:J\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7 \u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7'\\; . '.:: ..........0 6.51 9.51 : jj \/ JAPANESE OCTOPUS 10\u00c2\u00b728i\u00c2\u00a3 :_nc HOW SAY f B :rl4 :il: :wen2 t '- \u00e2\u0080\u00a2 :zhang!","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This alters generation of T as follows: n P (T |\u00cf\u0088) = n P (Ti|\u00cf\u0088) i=1 Note that this distribution captures the frequency of a tag across word types, as opposed to tokens.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Ends with the feminine affix :: p. 4.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"4.1 Corpora.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"When aligning the words in parallel texts (for language pairs like SpanishEnglish, French-English, ItalianGerman,...), we typically observe a strong localization effect.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"AdaBoost is given access to a weak learning algorithm, which accepts as input the training examples, along with a distribution over the instances.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Clearly this poses a number of research challenges, though, such as the applicability of tag sets across different languages.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The major problem for all segmentation systems remains the coverage afforded by the dictionary and the lexical rules used to augment the dictionary to deal with unseen words.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"\u00e2\u0080\u00a2 Some tools would allow for the desired annotation mode, but are so complicated (they can be used for many other purposes as well) that annotators take a long time getting used to them.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"If an unlabeled vertex does not have a path to any labeled vertex, this term ensures that the converged marginal for this vertex will be uniform over all tags, allowing the middle word of such an unlabeled vertex to take on any of the possible tags.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We present two algorithms.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Minimal perfect hashing is used to find the index at which a quantized probability and possibly backoff are stored.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"We believe that the underlying principles of the maximum entropy framework are suitable for exploiting information from diverse sources.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"One key issue here is to seek a discourse-based model of information structure.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Finally we show the combining techniques degrade very little when a poor parser is added to the set.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For pronouns, however, all of the knowledge sources increased recall, often substantially, and with little if any decrease in precision.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We would like to relax somewhat the constraint on the path complexity of formalisms in LCFRS.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In the CC-domain, there are 32 sets of phrases which contain more than 2 phrases.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This is less effective in our setting, where IN and OUT are disparate.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"f;g denotes the empty set, where no source sentence position is covered.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Given a document to process, BABAR uses four modules to perform coreference resolution.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Thus the method makes the fairly strong assumption that the features can be partitioned into two types such that each type alone is sufficient for classification.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"This can be seen as a rough approximation of Yarowsky and Ngai (2001).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We will also directly compare with a baseline similar to the Matsoukas et al approach in order to measure the benefit from weighting phrase pairs (or ngrams) rather than full sentences.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"But gold segmentation is not available in application settings, so a segmenter and parser are arranged in a pipeline.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We trained this model by optimizing the following objective function: Note that this involves marginalizing out all possible state configurations z for a sentence x, resulting in a non-convex objective.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Two aspects of the corpus have been presented in previous papers ((Re- itter, Stede 2003) on underspecified rhetorical structure; (Stede 2003) on the perspective of knowledge-based summarization).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In this specific case, as these two titles could fill the same column of an IE table, we regarded them as paraphrases for the evaluation.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"This design leads to a significant reduction in the computational complexity of training and inference.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The maximum likelihood estimates (i.e., parameter values which maximize 10) can not be found analytically, but the EM algorithm can be used to hill-climb to a local maximum of the likelihood function from some initial parameter settings.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Thus an explicit assumption about the redundancy of the features \u2014 that either the spelling or context alone should be sufficient to build a classifier \u2014 has been built into the algorithm.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Our analysis identifies three key factors driving our performance gain: 1) selecting a model structure which directly encodes tag sparsity, 2) a type-level prior on tag assignments, and 3) a straightforward na\u00c2\u00a8\u00c4\u00b1veBayes approach to incorporate features.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Again we notice that the isolated constituent precision is larger than 0.5 only in those partitions that contain very few samples.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"A less canonical representation of segmental morphology is triggered by a morpho-phonological process of omitting the definite article h when occurring after the particles b or l. This process triggers ambiguity as for the definiteness status of Nouns following these particles.We refer to such cases in which the concatenation of elements does not strictly correspond to the original surface form as super-segmental morphology.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Conditioned on T , features of word types W are drawn.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The 3rd block contains the mixture baselines.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Floating point values may be stored in the trie exactly, using 31 bits for non-positive log probability and 32 bits for backoff5.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"A greedy algorithm (or maximum-matching algorithm), GR: proceed through the sentence, taking the longest match with a dictionary entry at each point.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Comparison with state-of-the-art taggers For comparison we consider two unsupervised tag- gers: the HMM with log-linear features of Berg- Kirkpatrick et al.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"To explore this tradeoff, we have performed experiments with three different encoding schemes (plus a baseline), which are described schematically in Table 1.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In MUC6 and MUC7, the named entity task is defined as finding the following classes of names: person, organization, location, date, time, money, and percent (Chinchor, 1998; Sundheim, 1995) Machine learning systems in MUC6 and MUC 7 achieved accuracy comparable to rule-based systems on the named entity task.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Clearly the percentage of productively formed words is quite small (for this particular corpus), meaning that dictionary entries are covering most of the 15 GR is .73 or 96%..","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For example, the phrase \u00e2\u0080\u009c's New York-based trust unit,\u00e2\u0080\u009d is not a paraphrase of the other phrases in the \u00e2\u0080\u009cunit\u00e2\u0080\u009d set.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"paper, and is missing 6 examples from the A set.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"These conaUses lossy compression. bThe 8-bit quantized variant returned incorrect probabilities as explained in Section 3.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"This feature has a linguistic justification.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Riloff and Jones 99) was brought to our attention as we were preparing the final version of this paper.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The Recency KS computes the distance between the candidate and the anaphor relative to its scope.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"To quantize, we use the binning method (Federico and Bertoldi, 2006) that sorts values, divides into equally sized bins, and averages within each bin.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For 2 < n < N, we use a hash table mapping from the n-gram to the probability and backoff3.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"By sharing stacks (in IG's) or by using nonlinear equations over f-structures (in FUG's and LFG's), structures with unbounded dependencies between paths can be generated.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"2 56.2 32.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Alternatively, h can be thought of as defining a decision list of rules x y ranked by their &quot;strength&quot; h(x, y).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The record for wn1 stores the offset at which its extensions begin.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"We believe that the underlying principles of the maximum entropy framework are suitable for exploiting information from diverse sources.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Presence of the determiner J Al. 2.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The frequency of the Company \u00e2\u0080\u0093 Company domain ranks 11th with 35,567 examples.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We first define &quot;pseudo-labels&quot;,-yt, as follows: = Yi t sign(g 0\\ 2\u2014 kx2,m < i < n Thus the first m labels are simply copied from the labeled examples, while the remaining (n \u2014 m) examples are taken as the current output of the second classifier.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In the named entity domain these rules were Each of these rules was given a strength of 0.9999.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"A list of words occurring more than 10 times in the training data is also collected (commonWords).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"Here we present two algorithms.","target_text":"88,962 (spelling,context) pairs were extracted as training data.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"However, they list two sets, one consisting of 28 fragments and the other of 22 fragments, in which they had 0% recall and precision.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We are interested in combining the substructures of the input parses to produce a better parse.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"For one thing, it is not clear who is to receive settlements or what should happen in case not enough teachers accept the offer of early retirement.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"We can do that . IbmS: Yes, wonderful.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Any opinions, findings, conclusions, or recommendations expressed in this paper are those of the authors, and do not necessarily reflect the views of the funding organizations.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"A high-level relation is agent, which relates an animate nominal to a predicate.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"For instance, the sentence Similar improvements in haemoglobin levels were reported in the scientific literature for other epoetins would likely be considered domain-specific despite the presence of general phrases like were reported in.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses two methods to identify anaphors that can be easily and reliably resolved with their antecedent: lexical seeding and syntactic seeding.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"6 Joint Segmentation and Parsing.","label":0},{"source_text":"There is no global pruning.","target_text":"For the inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J), we drop the dependence on the target sentence length I. 2.2 Word Joining.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"E.g. when 'Zahnarzttermin' is aligned to dentist's, the extended lexicon model might learn that 'Zahnarzttermin' actuallyhas to be aligned to both dentist's and ap pointment.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, it is possible to personify any noun, so in children's stories or fables, i\u00c2\u00a5JJ1l.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"As expected, the most informative encoding, Head+Path, gives the highest accuracy with over 99% of all non-projective arcs being recovered correctly in both data sets.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Te rro ris m Na tur al Dis ast ers mu rde r of < NP > kill ed <p atie nt > <a ge nt > da ma ged wa s inj ure d in < NP > <a ge nt > rep ort ed <a ge nt > add ed <a ge nt > occ urr ed cau se of < NP > <a ge nt > stat ed <a ge nt > add ed <a ge nt > wr eak ed <a ge nt > cro sse d per pet rat ed <p atie nt > con de mn ed <p atie nt > dri ver of < NP > <a ge nt > car ryi ng Figure 1: Caseframe Network Examples Figure 1 shows examples of caseframes that co-occur in resolutions, both in the terrorism and natural disaster domains.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Frontier nodes are annotated by zero arty functions corresponding to elementary structures.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Model components cascade, so the row corresponding to +FEATS also includes the PRIOR component (see Section 3).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We use the log-linear tagger of Toutanova et al.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Given a sufficient number of randomly drawn unlabeled examples (i.e., edges), we will induce two completely connected components that together span the entire graph.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This technique was introduced by Clarkson and Rosenfeld (1997) and is also implemented by IRSTLM and BerkeleyLM\u2019s compressed option.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"A different notion of information structure, is used in work such as that of (?), who tried to characterize felicitous constituent ordering (theme choice, in particular) that leads to texts presenting information in a natural, \u00e2\u0080\u009cflowing\u00e2\u0080\u009d way rather than with abrupt shifts of attention.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"It is not immediately obvious how to formulate an equivalent to equation (1) for an adapted TM, because there is no well-defined objective for learning TMs from parallel corpora.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Clearly it is possible to write a rule that states that if an analysis Modal+ Verb is available, then that is to be preferred over Noun+ Verb: such a rule could be stated in terms of (finite-state) local grammars in the sense of Mohri (1993).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This class-based model gives reasonable results: for six radical classes, Table 1 gives the estimated cost for an unseen hanzi in the class occurring as the second hanzi in a double GIVEN name.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The portions of information in the large window can be individually clicked visible or invisible; here we have chosen to see (from top to bottom) \u00e2\u0080\u00a2 the full text, \u00e2\u0080\u00a2 the annotation values for the activated annotation set (co-reference), \u00e2\u0080\u00a2 the actual annotation tiers, and \u00e2\u0080\u00a2 the portion of text currently \u00e2\u0080\u0098in focus\u00e2\u0080\u0099 (which also appears underlined in the full text).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Suppose -y results from the adjunction of 71, ,-y, at the k distinct tree addresses 141, , nk in some elementary tree 7', respectively.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"With the additional assumption that (s, t) can be restricted to the support of co(s, t), this is equivalent to a \u201cflat\u201d alternative to (6) in which each non-zero co(s, t) is set to one.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In addition, the restricted version of CG's (discussed in Section 6) generates tree sets with independent paths and we hope that it can be included in a more general definition of LCFRS's containing formalisms whose tree sets have path sets that are themselves LCFRL's (as in the case of the restricted indexed grammars, and the hierarchy defined by Weir).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Roughly speaking, a language, L, has the property of semilinearity if the number of occurrences of each symbol in any string is a linear combination of the occurrences of these symbols in some fixed finite set of strings.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Let 71 be a tree with root and foot labeled by X.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"All commentaries have been annotated with rhetorical structure, using RSTTool4 and the definitions of discourse relations provided by Rhetorical Structure Theory (Mann, Thompson 1988).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"This is summarized in Equation 5.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Each derivation of a grammar can be represented by a generalized context-free derivation tree.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The following two error criteria are used in our experiments: mWER: multi-reference WER: We use the Levenshtein distance between the automatic translation and several reference translations as a measure of the translation errors.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Thus it is possible, for illustration, to look for a noun phrase (syntax tier) marked as topic (information structure tier) that is in a bridging relation (co-reference tier) to some other noun phrase.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"It also uses less memory, with 8 bytes of overhead per entry (we store 16-byte entries with m = 1.5); linked list implementations hash set and unordered require at least 8 bytes per entry for pointers.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The probability distribution that satisfies the above property is the one with the highest entropy.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The compressed variant uses block compression and is rather slow as a result.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"As we will see from Table 3, not much improvement is derived from this feature.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The algorithm can be viewed as heuristically optimizing an objective function suggested by (Blum and Mitchell 98); empirically it is shown to be quite successful in optimizing this criterion.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Second, rather than relying on a division of the corpus into manually-assigned portions, we use features intended to capture the usefulness of each phrase pair.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The weight on each sentence is a value in [0, 1] computed by a perceptron with Boolean features that indicate collection and genre membership.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We present two algorithms.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Personal names such as 00, 3R; zhoulenl-lai2 'Zhou Enlai.'","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Either save money at any cost - or give priority to education.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Next we show that the ATB is similar to other tree- banks in gross statistical terms, but that annotation consistency remains low relative to English (\u00c2\u00a73).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We are especially grateful to Taylor Berg- Kirkpatrick for running additional experiments.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"While sorted arrays could be used to implement the same data structure as PROBING, effectively making m = 1, we abandoned this implementation because it is slower and larger than a trie implementation.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We are currently exploring such algorithms.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This result suggests that all of contextual role KSs can provide useful information for resolving anaphora.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We focus on phrases which connect two Named Entities (NEs), and proceed in two stages.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This line of work has been motivated by empirical findings that the standard EM-learned unsupervised HMM does not exhibit sufficient word tag sparsity.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The path set of a tree set is the union of the path sets of trees in that tree set.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Limitations of (Blum and Mitchell 98): While the assumptions of (Blum and Mitchell 98) are useful in developing both theoretical results and an intuition for the problem, the assumptions are quite limited.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"\u2014 I would also like to point out to commissioner Liikanen that it is not easy to take a matter to a national court.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"When translating the sentence monotonically from left to right, the translation of the German finite verb 'kann', which is the left verbal brace in this case, is postponed until the German noun phrase 'mein Kollege' is translated, which is the subject of the sentence.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"We developed a first version of annotation guidelines for co-reference in PCC (Gross 2003), which served as basis for annotating the core corpus but have not been empirically evaluated for inter-annotator agreement yet.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"On the other hand, we are interested in the application of rhetorical analysis or \u00e2\u0080\u0098discourse parsing\u00e2\u0080\u0099 (3.2 and 3.3), in text generation (3.4), and in exploiting the corpus for the development of improved models of discourse structure (3.5).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If somewhere else in the document we see \u00e2\u0080\u009crestrictions put in place by President Bush\u00e2\u0080\u009d, then we can be surer that Bush is a name.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Our analysis begins with a description of syntactic ambiguity in unvocalized MSA text (\u00c2\u00a72).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"MCTAG's are able to generate tee sets having dependent paths.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Thus, most broad-coverage parsers based on dependency grammar have been restricted to projective structures.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In the present study, we limit ourselves to an algorithmic approach, using a deterministic breadthfirst search.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Thus the method makes the fairly strong assumption that the features can be partitioned into two types such that each type alone is sufficient for classification.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Our oracles took advantage of the labeled treebanks: While we tried to minimize the number of free parameters in our model, there are a few hyperparameters that need to be set.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For locations, tokens are matched against unigrams, and sequences of two consecutive tokens are matched against bigrams.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In our model there are no such hyper-parameters, and the performance is the result of truly joint disambiguation. sults.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"This is manifest in the lexical choices but 1 www.coli.unisb.de\/\u00e2\u0088\u00bcthorsten\/tnt\/ Dagmar Ziegler is up to her neck in debt.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For a sequence of hanzi that is a possible name, we wish to assign a probability to that sequence qua name.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"5 70.1 58.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The binary language model from Section 5.2 and text phrase table were forced into disk cache before each run.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"When a collision occurs, linear probing places the entry to be inserted in the next (higher index) empty bucket, wrapping around as necessary.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Cohen and Smith approach this by introducing the \u03b1 hyperparameter, which performs best when optimized independently for each sentence (cf.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"TPT Germann et al. (2009) describe tries with better locality properties, but did not release code.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The model was built with open vocabulary, modified Kneser-Ney smoothing, and default pruning settings that remove singletons of order 3 and higher.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"These words are in turn highly ambiguous, breaking the assumption underlying most parsers that the yield of a tree for a given sentence is known in advance.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Because we are interested in applying our techniques to languages for which no labeled resources are available, we paid particular attention to minimize the number of free parameters and used the same hyperparameters for all language pairs.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In IE, creating the patterns which express the requested scenario, e.g. \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d or \u00e2\u0080\u009ccorporate merger and acquisition\u00e2\u0080\u009d is regarded as the hardest task.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In sequential tagging models such as (Adler and Elhadad, 2006; Bar-Haim et al., 2007; Smith et al., 2005) weights are assigned according to a language model The input for the joint task is a sequence W = w1, ... , wn of space-delimited tokens.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The improvement is due to the cost of bit-level reads and avoiding reads that may fall in different virtual memory pages.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In total, for the 2,000 NE category pairs, 5,184 keywords are found.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Ties are rare in Bayes switching because the models are fine-grained \u2014 many estimated probabilities are involved in each decision.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"During the search process, a partial hypothesis is extended by choosing a source sentence position, which has not been aligned with a target sentence position yet.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"By contrast, when we turn to a comparison of the three encoding schemes it is hard to find any significant differences, and the overall impression is that it makes little or no difference which encoding scheme is used, as long as there is some indication of which words are assigned their linear head instead of their syntactic head by the projective parser.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"More recently, Subramanya et al. (2010) defined a graph over the cliques in an underlying structured prediction model.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This suggests that the backoff model is as reasonable a model as we can use in the absence of further information about the expected cost of a plural form.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Each decision determines the inclusion or exclusion of a candidate constituent.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For a language like English, this problem is generally regarded as trivial since words are delimited in English text by whitespace or marks of punctuation.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Our results outperform strong unsupervised baselines as well as approaches that rely on direct projections, and bridge the gap between purely supervised and unsupervised POS tagging models.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In general, different modalities (\u00e2\u0080\u009cplanned to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cagreed to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cbought\u00e2\u0080\u009d) were considered to express the same relationship within an extraction setting.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Such resources exist for Hebrew (Itai et al., 2006), but unfortunately use a tagging scheme which is incompatible with the one of the Hebrew Treebank.s For this reason, we use a data-driven morphological analyzer derived from the training data similar to (Cohen and Smith, 2007).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Our method does not assume any knowledge about the target language (in particular no tagging dictionary is assumed), making it applicable to a wide array of resource-poor languages.","label":1},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We evaluate the time and memory consumption of each data structure by computing perplexity on 4 billion tokens from the English Gigaword corpus (Parker et al., 2009).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Our second point of comparison is with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"rhetorical analysis We are experimenting with a hybrid statistical and knowledge-based system for discourse parsing and summarization (Stede 2003), (Hanneforth et al. 2003), again targeting the genre of commentaries.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"1 53.8 47.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Projectivizing a dependency graph by lifting nonprojective arcs is a nondeterministic operation in the general case.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Sparse lookup is a key subproblem of language model queries.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Presence of the determiner J Al. 2.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The results clearly indicate increased variation in the ATB relative to the WSJ, but care should be taken in assessing the magnitude of the difference.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":", for A. T.&T. nonalpha.. .","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The additional morphological material in such cases appears after the stem and realizes the extended meaning.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We experimented with increasingly rich grammars read off of the treebank.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Test set OOV rate is computed using the following splits: ATB (Chiang et al., 2006); CTB6 (Huang and Harper, 2009); Negra (Dubey and Keller, 2003); English, sections 221 (train) and section 23 (test).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Note that in our formalism a weakhypothesis can abstain.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Like HG's, TAG's, and MCTAG's, members of LCFRS can manipulate structures more complex than terminal strings and use composition operations that are more complex that concatenation.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Here, annotation proceeds in two phases: first, the domains and the units of IS are marked as such.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The model was built with open vocabulary, modified Kneser-Ney smoothing, and default pruning settings that remove singletons of order 3 and higher.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"A better approach would be to distin guish between these cases, possibly by drawing on the vast linguistic work on Arabic connectives (AlBatal, 1990).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For the disasters domain, 8245 texts were used for training and the 40 test documents contained 447 anaphoric links.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"On the other hand, if it is seen as McCann Pte.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"(1998) have also used a maximum entropy classifier that uses already tagged entities to help tag other entities.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"With a human evaluation we also showed that ATB inter-annotator agreement remains low relative to the WSJ corpus.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"SRILM\u2019s compact variant has an incredibly expensive destructor, dwarfing the time it takes to perform translation, and so we also modified Moses to avoiding the destructor by calling exit instead of returning normally.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For all lists except locations, the lists are processed into a list of tokens (unigrams).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The terms on the right-hand-side denote the type-level and token-level probability terms respectively.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"We then evaluate the approach in two steps.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The precision and recall measures (described in more detail in Section 3) used in evaluating Treebank parsing treat each constituent as a separate entity, a minimal unit of correctness.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Thus at each iteration the algorithm is forced to pick features for the location, person and organization in turn for the classifier being trained.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"A very small excerpt from an Italian-English graph is shown in Figure 1.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"KS Function Ge nde r filters candidate if gender doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We will briefly discuss this point in Section 3.1.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"2 56.2 32.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"An IG can be viewed as a CFG in which each nonterminal is associated with a stack.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Next, we represent the input sentence as an unweighted finite-state acceptor (FSA) I over H. Let us assume the existence of a function Id, which takes as input an FSA A, and produces as output a transducer that maps all and only the strings of symbols accepted by A to themselves (Kaplan and Kay 1994).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"As is standard, we use a fixed constant K for the number of tagging states.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Arguably this consists of about three phonological words.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Due to the dramatic fiscal situation in Brandenburg she now surprisingly withdrew legislation drafted more than a year ago, and suggested to decide on it not before 2003.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"With a good hash function, collisions of the full 64bit hash are exceedingly rare: one in 266 billion queries for our baseline model will falsely find a key not present.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"For each 13 Of course, this weighting makes the PCFG an improper distribution.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"A moment's reflection will reveal that things are not quite that simple.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"First, we learn weights on individual phrase pairs rather than sentences.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We generate these caseframes automatically by running AutoSlog over the training corpus exhaustively so that it literally generates a pattern to extract every noun phrase in the corpus.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We are unaware of prior results for the Stanford parser.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"However, the next step is clearly different.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In principle, it would be possible to encode the exact position of the syntactic head in the label of the arc from the linear head, but this would give a potentially infinite set of arc labels and would make the training of the parser very hard.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Thus it is possible, for illustration, to look for a noun phrase (syntax tier) marked as topic (information structure tier) that is in a bridging relation (co-reference tier) to some other noun phrase.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Mi(c) is a binary function returning t when parser i (from among the k parsers) suggests constituent c should be in the parse.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The breakdown of the different types of words found by ST in the test corpus is given in Table 3.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The use of ILP in learning the desired grammar significantly increases the computational complexity of this method.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The following recursive equation is evaluated: Qe0 (e; S; C; j) = (2) = p(fj je) max \u00c3\u0086;e00 np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) max (S0;j0) (S0 ;Cnfjg;j0)!(S;C;j) j02Cnfjg Qe00 (e0; S0;C n fjg; j0)o: The search ends in the hypotheses (I; f1; ; Jg; j).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"9 65.5 46.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Before describing the unsupervised case we first describe the supervised version of the algorithm: Input to the learning algorithm: n labeled examples of the form (xi, y\u201e). y, is the label of the ith example (given that there are k possible labels, y, is a member of y = {1 ... 0). xi is a set of mi features {x,1, Xi2 .","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Hash tables are a common sparse mapping technique used by SRILM\u2019s default and BerkeleyLM\u2019s hashed variant.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The toplevel weights are trained to maximize a metric such as BLEU on a small development set of approximately 1000 sentence pairs.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"There are clearly eight orthographic words in the example given, but if one were doing syntactic analysis one would probably want to consider I'm to consist of two syntactic words, namely I and am.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The computing time is low, since no reordering is carried out.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This essentially means that a better grammar tunes the joint model for optimized syntactic disambiguation at least in as much as their hyper parameters do.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We focus here instead on adapting the two most important features: the language model (LM), which estimates the probability p(wIh) of a target word w following an ngram h; and the translation models (TM) p(slt) and p(t1s), which give the probability of source phrase s translating to target phrase t, and vice versa.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"(b) After they were released...","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The BerkeleyLM direct-mapped cache is in principle faster than caches implemented by RandLM and by IRSTLM, so we may write a C++ equivalent implementation as future work.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Clearly this poses a number of research challenges, though, such as the applicability of tag sets across different languages.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Multiple features can be used for the same token.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"8 57.3 +F EA TS be st me dia n 50.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"As mentioned above, it is not obvious how to apply Daum\u00b4e\u2019s approach to multinomials, which do not have a mechanism for combining split features.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The second model (+PRIOR) utilizes the independent prior over type-level tag assignments P (T |\u00cf\u0088).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"It is also true of the adaptation of the Collins parser for Czech (Collins et al., 1999) and the finite-state dependency parser for Turkish by Oflazer (2003).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We allow any number of bits from 2 to 25, unlike IRSTLM (8 bits) and BerkeleyLM (17\u221220 bits).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"73 81.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Having explained the various layers of annotation in PCC, we now turn to the question what all this might be good for.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The system of Berg-Kirkpatrick et al.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The details of the transformation procedure are slightly different depending on the encoding schemes: d\u2191h let the linear head be the syntactic head). target arc must have the form wl \u2212\u2192 wm; if no target arc is found, Head is used as backoff. must have the form wl \u2212\u2192 wm and no outgoing arcs of the form wm p'\u2193 \u2212\u2192 wo; no backoff.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The iw all map variant uses a non-0 y weight on a uniform prior in p,,(s t), and outperforms a version with y = 0 (iw all) and the \u201cflattened\u201d variant described in section 3.2.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"37.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Word Re-ordering and DP-based Search in Statistical Machine Translation","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"A position is presented by the word at that position.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"am 11.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This suggests that the backoff model is as reasonable a model as we can use in the absence of further information about the expected cost of a plural form.","label":0},{"source_text":"There is no global pruning.","target_text":"Then, the German infinitive 'besuchen' and the negation particle 'nicht' are translated.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"However, 1 http:\/\/maxent.sourceforge.net 3.2 Testing.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"21 84.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The work reported here is closely related to [Ha- segawa et al. 04].","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"4.5 Transliterations of Foreign Words.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Here, an NE instance pair is any pair of NEs separated by at most 4 syntactic chunks; for example, \u00e2\u0080\u009cIBM plans to acquire Lotus\u00e2\u0080\u009d.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"At each iteration the algorithm increases the number of rules, while maintaining a high level of agreement between the spelling and contextual decision lists.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"Here we present two algorithms.","target_text":"There are two differences between this method and the DL-CoTrain algorithm: spelling and contextual features, alternating between labeling and learning with the two types of features.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We define an ATM, M, recognizing a language generated by a grammar, G, having the properties discussed in Section 43.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We incur some additional memory cost due to storing state in each hypothesis, though this is minimal compared with the size of the model itself.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For each source word f, the list of its possible translations e is sorted according to p(fje) puni(e), where puni(e) is the unigram probability of the English word e. It is su\u00c3\u0086cient to consider only the best 50 words.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Finally, we thank Kuzman Ganchev and the three anonymous reviewers for helpful suggestions and comments on earlier drafts of this paper.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The only way to handle such phenomena within the framework described here is simply to expand out the reduplicated forms beforehand, and incorporate the expanded forms into the lexical transducer.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Daum\u00b4e (2007) applies a related idea in a simpler way, by splitting features into general and domain-specific versions.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"At the phrasal level, we remove all function tags and traces.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We compare the manually annotated grammar, which we incorporate into the Stanford parser, to both the Berkeley (Petrov et al., 2006) and Bikel (Bikel, 2004) parsers.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Our results suggest that it is possible to learn accurate POS taggers for languages which do not have any annotated data, but have translations into a resource-rich language.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"There has been a lot of research on such lexical relations, along with the creation of resources such as WordNet.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"All experiments use ATB parts 1\u00e2\u0080\u00933 divided according to the canonical split suggested by Chiang et al.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Several systems propose statistical methods for handling unknown words (Chang et al. 1992; Lin, Chiang, and Su 1993; Peng and Chang 1993).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"information structure.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"When using the segmentation pruning (using HSPELL) for unseen tokens, performance improves for all tasks as well.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"2) An improved language model, which takes into account syntactic structure, e.g. to ensure that a proper English verbgroup is generated.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"La ng ua ge # To ke ns # W or d Ty pe s # Ta gs E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 1 1 7 3 7 6 6 9 4 3 8 6 2 0 3 5 6 8 6 9 9 6 0 5 2 0 6 6 7 8 8 9 3 3 4 1 9 1 4 6 7 4 9 2 0 6 1 8 3 5 6 2 8 3 9 3 7 2 3 2 5 2 8 9 3 1 1 6 4 5 8 2 0 0 5 7 4 5 2 5 1 2 5 4 2 2 4 7 4 1 Table 2: Statistics for various corpora utilized in experiments.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We have presented two general approaches to studying parser combination: parser switching and parse hybridization.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We propose a limit of 70 words for Arabic parsing evaluations.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Since the inclusion of out-ofdomain test data was a very late decision, the participants were not informed of this.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The first shows how constituent features and context do not help in deciding which parser to trust.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Although the ultimate goal of the Verbmobil project is the translation of spoken language, the input used for the translation experiments reported on in this paper is the (more or less) correct orthographic transcription of the spoken sentences.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Precision is the portion of hypothesized constituents that are correct and recall is the portion of the Treebank constituents that are hypothesized.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"This vector tx is constructed for every word in the foreign vocabulary and will be used to provide features for the unsupervised foreign language POS tagger.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Out of those 15 links, 4 are errors, namely \u00e2\u0080\u009cbuy - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cpurchase - stake\u00e2\u0080\u009d \u00e2\u0080\u009cacquisition - stake\u00e2\u0080\u009d.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For example, suppose one is building a ITS system for Mandarin Chinese.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Apart from MERT difficulties, a conceptual problem with log-linear combination is that it multiplies feature probabilities, essentially forcing different features to agree on high-scoring candidates.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We can 5 Recall that precision is defined to be the number of correct hits divided by the total number of items.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":1},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This is to allow for fair comparison between the statistical method and GR, which is also purely dictionary-based.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The code is opensource, has minimal dependencies, and offers both C++ and Java interfaces for integration.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"First, the model assumes independence between the first and second hanzi of a double given name.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We group the features used into feature groups.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"0 57.2 43.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The authors provided us with a ratio between TPT and SRI under different conditions. aLossy compression with the same weights. bLossy compression with retuned weights. ditions make the value appropriate for estimating repeated run times, such as in parameter tuning.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"If the log backoff of wnf is also zero (it may not be in filtered models), then wf should be omitted from the state.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"In this case we are interested in finding' the maximum probability parse, ri, and Mi is the set of relevant (binary) parsing decisions made by parser i. ri is a parse selected from among the outputs of the individual parsers.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Second, we show that although the Penn Arabic Treebank is similar to other tree- banks in gross statistical terms, annotation consistency remains problematic.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Email: gale@research.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"On several languages, we report performance exceeding that of more complex state-of-the art systems.1","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"11 www.ling.unipotsdam.de\/sfb\/projekt a3.php 12 This step was carried out in the course of the diploma thesis work of David Reitter (2003), which de serves special mention here.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"But for eign learners are often surprised by the verbless predications that are frequently used in Arabic.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"All our results are obtained by using only the official training data provided by the MUC conferences.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"2.1.1 Lexical Seeding It is generally not safe to assume that multiple occurrences of a noun phrase refer to the same entity.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"If (wi, r, wj) E A, we say that wi is the head of wj and wj a dependent of wi.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"It is also true of the adaptation of the Collins parser for Czech (Collins et al., 1999) and the finite-state dependency parser for Turkish by Oflazer (2003).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Note that since all English vertices were extracted from the parallel text, we will have an initial label distribution for all vertices in Ve.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"2.2 Syntactic structure.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We have shown that, at least given independent human judgments, this is not the case, and that therefore such simplistic measures should be mistrusted.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We tried two versions of our graph-based approach: feature after the first stage of label propagation (Eq.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Values in the trie are minimally sized at the bit level, improving memory consumption over trie implementations in SRILM, IRSTLM, and BerkeleyLM.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The domains are the linguistic spans that are to receive an IS-partitioning, and the units are the (smaller) spans that can play a role as a constituent of such a partitioning.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"In Table 7 we give results for several evaluation metrics.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Manual annotation results in human in- terpretable grammars that can inform future tree- bank annotation decisions.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Unsupervised Part-of-Speech Tagging with Bilingual Graph-Based Projections","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For the inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J), we drop the dependence on the target sentence length I. 2.2 Word Joining.","label":0},{"source_text":"There is no global pruning.","target_text":"On the other hand, only very restricted reorderings are necessary, e.g. for the translation direction from Table 2: Coverage set hypothesis extensions for the IBM reordering.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"There is a fairly large body of work on SMT adaptation.","label":0},{"source_text":"There is no global pruning.","target_text":"(1), Pr(eI 1) is the language model, which is a trigram language model in this case.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"In the small window on the left, search queries can be entered, here one for an NP that has been annotated on the co-reference layer as bridging.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Specifically, we assume each word type W consists of feature-value pairs (f, v).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Like in the co-reference annotation, G\u00c2\u00a8otze\u00e2\u0080\u0099s proposal has been applied by two annotators to the core corpus but it has not been systematically evaluated yet.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"It is well-known that constituency parsing models designed for English often do not generalize easily to other languages and treebanks.1 Explanations for this phenomenon have included the relative informativeness of lexicalization (Dubey and Keller, 2003; Arun and Keller, 2005), insensitivity to morphology (Cowan and Collins, 2005; Tsarfaty and Sima\u00e2\u0080\u0099an, 2008), and the effect of variable word order (Collins et al., 1999).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Finally, we model the probability of a new transliterated name as the product of PTN and PTN(hanzi;) for each hanzi; in the putative name.13 The foreign name model is implemented as an WFST, which is then summed with the WFST implementing the dictionary, morpho 13 The current model is too simplistic in several respects.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The dictionary sizes reported in the literature range from 17,000 to 125,000 entries, and it seems reasonable to assume that the coverage of the base dictionary constitutes a major factor in the performance of the various approaches, possibly more important than the particular set of methods used in the segmentation.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As described in Sproat (1995), the Chinese segmenter presented here fits directly into the context of a broader finite-state model of text analysis for speech synthesis.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Additionally, it works about 3 times as fast as the IBM style search.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Turning now to (1), we have the similar problem that splitting.into.ma3 'horse' andlu4 'way' is more costly than retaining this as one word .ma3lu4 'road.'","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Evalb is a Java re-implementation of the standard labeled precision\/recall metric.12 The ATB gives all punctuation a single tag.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Finally, the focus\/background partition is annotated, together with the focus question that elicits the corresponding answer.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Our analysis begins with a description of syntactic ambiguity in unvocalized MSA text (\u00c2\u00a72).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Even this may be nondeterministic, in case the graph contains several non-projective arcs whose lifts interact, but we use the following algorithm to construct a minimal projective transformation D0 = (W, A0) of a (nonprojective) dependency graph D = (W, A): The function SMALLEST-NONP-ARC returns the non-projective arc with the shortest distance from head to dependent (breaking ties from left to right).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Lemma: If the number of votes required by constituent voting is greater than half of the parsers under consideration the resulting structure has no crossing constituents.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The percentage scores on the axis labels represent the amount of variation in the data explained by the dimension in question.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The next two rows are results of oracle experiments.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We can only compare with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Again, famous place names will most likely be found in the dictionary, but less well-known names, such as 1PM\u00c2\u00b1 R; bu4lang3-shi4wei2-ke4 'Brunswick' (as in the New Jersey town name 'New Brunswick') will not generally be found.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The P (T |\u00cf\u0088) distribution, in English for instance, should have very low mass for the DT (determiner) tag, since determiners are a very small portion of the vocabulary.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Some relations are signalled by subordinating conjunctions, which clearly demarcate the range of the text spans related (matrix clause, embedded clause).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Source sentence words are aligned with hypothesized target sentence words, where the choice of a new source word, which has not been aligned with a target word yet, is restricted1.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Using less training data than other systems, our NER is able to perform as well as other state-of-the-art NERs.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We describe a novel approach for inducing unsupervised part-of-speech taggers for languages that have no labeled training data, but have translated text in a resource-rich language.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This section measures performance on shared tasks in order of increasing complexity: sparse lookups, evaluating perplexity of a large file, and translation with Moses.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The final block in table 2 shows models trained on feature subsets and on the SVM feature described in 3.4.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"This drastic tree manipulation is not appropriate for situations in which we want to assign particular structures to sentences.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Two lists, Corporate-Suffix-List (for corporate suffixes) and Person-Prefix-List (for person prefixes), are collected from the training data.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"On the MUC6 data, Bikel et al.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Table 3 contains the results for evaluating our systems on the test set (section 22).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We call such a constituent an isolated constituent.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The problem is a binary classification problem.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Note that hanzi that are not grouped into dictionary words (and are not identified as single\u00c2\u00ad hanzi words), or into one of the other categories of words discussed in this paper, are left unattached and tagged as unknown words.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This is not unreasonable given the application to phrase pairs from OUT, but it suggests that an interesting alternative might be to use a plain log-linear weighting function exp(Ei Aifi(s, t)), with outputs in [0, oo].","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The state function is integrated into the query process so that, in lieu of the query p(wnjwn\u22121 1 ), the application issues query p(wnjs(wn\u22121 1 )) which also returns s(wn1 ).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"While Gan's system incorporates fairly sophisticated models of various linguistic information, it has the drawback that it has only been tested with a very small lexicon (a few hundred words) and on a very small test set (thirty sentences); there is therefore serious concern as to whether the methods that he discusses are scalable.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Rather we believe several methods have to be developed using different heuristics to discover wider variety of paraphrases.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Finally, we note that simple weighting gives nearly a 2% F1 improvement, whereas Goldberg and Tsarfaty (2008) found that unweighted lattices were more effective for Hebrew.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Interpolation search formalizes the notion that one opens a dictionary near the end to find the word \u201czebra.\u201d Initially, the algorithm knows the array begins at b +\u2014 0 and ends at e +\u2014 |A|\u22121.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Se ma nti c (a) filters candidate if its semantic tags d o n \u00e2\u0080\u0099 t i n t e r s e c t w i t h t h o s e o f t h e a n a p h o r .","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In (b) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the kidnapping victims, but in (c) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the armed men.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Pumping t2 will change only one branch and leave the other branch unaffected.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The method uses a &quot;soft&quot; measure of the agreement between two classifiers as an objective function; we described an algorithm which directly optimizes this function.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"For each language under consideration, Petrov et al. (2011) provide a mapping A from the fine-grained language specific POS tags in the foreign treebank to the universal POS tags.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The maximum entropy framework estimates probabilities based on the principle of making as few assumptions as possible, other than the constraints imposed.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"For each token, three types of features may be taken into account: the word form; the part-of-speech assigned by an automatic tagger; and labels on previously assigned dependency arcs involving the token \u2013 the arc from its head and the arcs to its leftmost and rightmost dependent, respectively.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Following Dickinson (2005), we randomly sampled 100 variation nuclei from each corpus and evaluated each sample for the presence of an annotation error.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The most accurate characterization of Chinese writing is that it is morphosyllabic (DeFrancis 1984): each hanzi represents one morpheme lexically and semantically, and one syllable phonologi\u00c2\u00ad cally.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Log-linear combination (loglin) improves on this in all cases, and also beats the pure IN system.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Its success depends on the two domains being relatively close, and on the OUT corpus not being so large as to overwhelm the contribution of IN.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"3.2 Results.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"newspaper material, but also including kungfu fiction, Buddhist tracts, and scientific material.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For our demonstration system, we typically use the pruning threshold t0 = 5:0 to speed up the search by a factor 5 while allowing for a small degradation in translation accuracy.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In the named entity domain these rules were Each of these rules was given a strength of 0.9999.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"For \u00c3\u0086 = 0, no new target word is generated, while an additional source sentence position is covered.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.3 Rhetorical structure.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Mai.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The last affix in the list is the nominal plural f, men0.20 In the table are the (typical) classes of words to which the affix attaches, the number found in the test corpus by the method, the number correct (with a precision measure), and the number missed (with a recall measure).","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"(Brandt 1996) extended these ideas toward a conception of kommunikative Gewichtung (\u00e2\u0080\u0098communicative-weight assignment\u00e2\u0080\u0099).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The constituent voting and na\u00efve Bayes techniques are equivalent because the parameters learned in the training set did not sufficiently discriminate between the three parsers.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Process nominals name the action of the transitive or ditransitive verb from which they derive.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Third, all remaining anaphora are evaluated by 11 different knowledge sources: the four contextual role knowledge sources just described and seven general knowledge sources.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The joint morphological and syntactic hypothesis was first discussed in (Tsarfaty, 2006; Tsarfaty and Sima\u2019an, 2004) and empirically explored in (Tsarfaty, 2006).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Our first model is GTplain, a PCFG learned from the treebank after removing all functional features from the syntactic categories.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Clearly, retaining the original frequencies is important for good performance, and globally smoothing the final weighted frequencies is crucial.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Tables 4 and 5 also show that putting all of the contextual role KSs in play at the same time produces the greatest performance gain.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The effect of the pruning threshold t0 is shown in Table 5.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"For all languages, the vocabulary sizes increase by several thousand words.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For example, it is well-known that one can build a finite-state bigram (word) model by simply assigning a state Si to each word Wi in the vocabulary, and having (word) arcs leaving that state weighted such that for each Wj and corresponding arc aj leaving Si, the cost on aj is the bigram cost of WiWj- (Costs for unseen bigrams in such a scheme would typically be modeled with a special backoff state.)","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For example, the independence assumptions mean that the model fails to capture the dependence between specific and more general features (for example the fact that the feature full.-string=New_York is always seen with the features contains (New) and The baseline method tags all entities as the most frequent class type (organization). contains (York) and is never seen with a feature such as contains (Group) ).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In order to observe the similarity between these constrained systems, it is crucial to abstract away from the details of the structures and operations used by the system.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For this example, since the sequence Even News Broadcasting Corp. only appears once in the document, its longest substring that occurs in the same document is News Broadcasting Corp. In this case, News has an additional feature of I begin set to 1, Broadcasting has an additional feature of I continue set to 1, and Corp. has an additional feature of I end set to 1.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The idea is to have a pipeline of shallow-analysis modules (tagging, chunk- ing, discourse parsing based on connectives) and map the resulting underspecified rhetorical tree (see Section 2.4) into a knowledge base that may contain domain and world knowledge for enriching the representation, e.g., to resolve references that cannot be handled by shallow methods, or to hypothesize coherence relations.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"A name already mentioned previously in a document may appear in abbreviated form when it is mentioned again later.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"All experiments use ATB parts 1\u00e2\u0080\u00933 divided according to the canonical split suggested by Chiang et al.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In this paper, we outline how such family of formalisms can be defined, and show that like CFG's, each member possesses a number of desirable linguistic and computational properties: in particular, the constant growth property and polynomial recognizability.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"These universal POS categories not only facilitate the transfer of POS information from one language to another, but also relieve us from using controversial evaluation metrics,2 by establishing a direct correspondence between the induced hidden states in the foreign language and the observed English labels.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The search starts in the hypothesis (I; f;g; 0).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"For the remaining arcs, if the segment is in fact a known lexeme it is tagged as usual, but for the OOV arcs which are valid Hebrew entries lacking tags assignment, we assign all possible tags and then simulate a grammatical constraint.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In the experiments below, we employ a data-driven deterministic dependency parser producing labeled projective dependency graphs,3 previously tested on Swedish (Nivre et al., 2004) and English (Nivre and Scholz, 2004).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Another attempt at using global information can be found in (Borthwick, 1999).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"2 Chinese ?l* han4zi4 'Chinese character'; this is the same word as Japanese kanji..","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Different sentence structure and rich target language morphology are two reasons for this.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In many cases, inspection of either the spelling or context alone is sufficient to classify an example.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"There is no global pruning.","target_text":"10.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The global feature groups are: InitCaps of Other Occurrences (ICOC): There are 2 features in this group, checking for whether the first occurrence of the same word in an unambiguous position (non first-words in the TXT or TEXT zones) in the same document is initCaps or not-initCaps.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"For effectively annotating connectives\/scopes, we found that existing annotation tools were not well-suited, for two reasons: \u00e2\u0080\u00a2 Some tools are dedicated to modes of annotation (e.g., tiers), which could only quite un-intuitively be used for connectives and scopes.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"For each experiment we gave an nonparametric and a parametric technique for combining parsers.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"This means that the PCC cannot grow particularly quickly.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Since the early days of statistical NLP, researchers have observed that a part-of-speech tag distribution exhibits \u00e2\u0080\u009cone tag per discourse\u00e2\u0080\u009d sparsity \u00e2\u0080\u0094 words are likely to select a single predominant tag in a corpus, even when several tags are possible.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"However, there is a strong relationship between ni1s and the number of hanzi in the class.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Obviously \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is part of the following clause rather than being the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d and the extracted instance makes no sense.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Having found (spelling, context) pairs in the parsed data, a number of features are extracted.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"html 4 www.wagsoft.com\/RSTTool assigning rhetorical relations is a process loaded with ambiguity and, possibly, subjectivity.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"First, we use a novel graph-based framework for projecting syntactic information across language boundaries.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Roughly speaking, the new algorithm presented in this paper performs a similar search, but instead minimizes a bound on the number of (unlabeled) examples on which two classifiers disagree.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold, 30.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"0 D\/ nc 5.0 The minimal dictionary encoding this information is represented by the WFST in Figure 2(a).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Alternatively, h can be thought of as defining a decision list of rules x y ranked by their &quot;strength&quot; h(x, y).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Figure 2 shows timing results.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Based on revision 4041, we modified Moses to print process statistics before terminating.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The corpus was wordaligned using both HMM and IBM2 models, and the phrase table was the union of phrases extracted from these separate alignments, with a length limit of 7.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"When aligning the words in parallel texts (for language pairs like SpanishEnglish, French-English, ItalianGerman,...), we typically observe a strong localization effect.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We use a squared loss to penalize neighboring vertices that have different label distributions: kqi \u2212 qjk2 = Ey(qi(y) \u2212 qj(y))2, and additionally regularize the label distributions towards the uniform distribution U over all possible labels Y.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"For comparison to information-retrieval inspired baselines, eg (L\u00a8u et al., 2007), we select sentences from OUT using language model perplexities from IN.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Applying the function PROJECTIVIZE to the graph in Figure 1 yields the graph in Figure 2, where the problematic arc pointing to Z has been lifted from the original head jedna to the ancestor je.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This information is readily available in TRIE where adjacent records with equal pointers indicate no further extension of context is possible.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Pr(eI 1) is the language model of the target language, whereas Pr(fJ 1 jeI1) is the transla tion model.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"There are two weaknesses in Chang et al.'s model, which we improve upon.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Jiang and Zhai (2007) suggest the following derivation, making use of the true OUT distribution po(s, t): where each fi(s, t) is a feature intended to charac- !0\u02c6 = argmax pf(s, t) log p\u03b8(s|t) (8) terize the usefulness of (s, t), weighted by Ai. \u03b8 s,t pf(s, t)po(s, t) log p\u03b8(s|t) The mixing parameters and feature weights (col- != argmax po (s, t) lectively 0) are optimized simultaneously using dev- \u03b8 s,t pf(s, t)co(s, t) log p\u03b8(s|t), set maximum likelihood as before: !\ufffdargmax po (s, t) ! \u03b8 s,t \ufffd\u02c6 = argmax \u02dcp(s, t) log p(s|t; 0).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"These estimates are in turn combined linearly with relative-frequency estimates from an in-domain phrase table.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Dagan and Itai (Dagan and Itai, 1990) experimented with co-occurrence statistics that are similar to our lexical caseframe expectations.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Each vertex within a connected component must have the same label \u2014 in the binary classification case, we need a single labeled example to identify which component should get which label.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A minimal requirement for building a Chinese word segmenter is obviously a dictionary; furthermore, as has been argued persuasively by Fung and Wu (1994), one will perform much better at segmenting text by using a dictionary constructed with text of the same genre as the text to be segmented.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In these examples, the names identified by the two systems (if any) are underlined; the sentence with the correct segmentation is boxed.19 The differences in performance between the two systems relate directly to three issues, which can be seen as differences in the tuning of the models, rather than repre\u00c2\u00ad senting differences in the capabilities of the model per se.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"RandLM\u2019s stupid backoff variant stores counts instead of probabilities and backoffs.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Top 5 Bot to m 5 Go ld NN P NN JJ CD NN S RB S PD T # \u00e2\u0080\u009d , 1T W CD W RB NN S VB N NN PR P$ W DT : MD . +P RI OR CD JJ NN S WP $ NN RR B- , $ \u00e2\u0080\u009d . +F EA TS JJ NN S CD NN P UH , PR P$ # . \u00e2\u0080\u009c Table 5: Type-level English POS Tag Ranking: We list the top 5 and bottom 5 POS tags in the lexicon and the predictions of our models under the best hyperparameter setting.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"For even larger models, we recommend RandLM; the memory consumption of the cache is not expected to grow with model size, and it has been reported to scale well.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The experimental tests are carried out on the Verbmobil task (GermanEnglish, 8000-word vocabulary), which is a limited-domain spoken-language task.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We can better predict the probability of an unseen hanzi occurring in a name by computing a within-class Good-Turing estimate for each radical class.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"First, we use a novel graph-based framework for projecting syntactic information across language boundaries.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The corpus has been annotated with six different types of information, which are characterized in the following subsections.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Previous approaches have tried to find examples that are similar to the target domain.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Machine Translation In this case my colleague can not visit on I n d i e s e m F a l l ka nn m e i n K o l l e g e a m the v i e r t e n M a i n i c h t b e s u c h e n S i e you fourth of May Figure 1: Reordering for the German verbgroup.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Previous work on morphological and syntactic disambiguation in Hebrew used different sets of data, different splits, differing annotation schemes, and different evaluation measures.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.3 Assigning Evidence Values.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We create equivalence classes for verb, noun, and adjective POS categories.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"This is equivalent to the assumption used in probability estimation for na\u00efve Bayes classifiers, namely that the attribute values are conditionally independent when the target value is given.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Here a set is represented by the keyword and the number in parentheses indicates the number of shared NE pair instances.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The ATB annotation distinguishes between verbal and nominal readings of maSdar process nominals.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"However, when grammatical relations like subject and object are evaluated, parsing performance drops considerably (Green et al., 2009).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"When SRILM estimates a model, it sometimes removes n-grams but not n + 1-grams that extend it to the left.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Finally, we make some improvements to baseline approaches.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Having explained the various layers of annotation in PCC, we now turn to the question what all this might be good for.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"However, there are phrases which express the same meanings even though they do not share the same keyword.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"This pumping lemma states that if there is tree, t = t2t3t4t5, generated by a TAG G, such that its height is more than a predetermined bound k, then all trees of the form ti it tstt ts for each i > 0 will also generated by G. Similarly, for tree sets with independent paths and more complex path sets, tree pumping lemmas can be given.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"To see this, note thai the first two terms in the above equation correspond to the function that AdaBoost attempts to minimize in the standard supervised setting (Equ.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Preprocessing the raw trees improves parsing performance considerably.9 We first discard all trees dominated by X, which indicates errors and non-linguistic text.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Feature weights were set using Och\u2019s MERT algorithm (Och, 2003).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"This is manifest in the lexical choices but 1 www.coli.unisb.de\/\u00e2\u0088\u00bcthorsten\/tnt\/ Dagmar Ziegler is up to her neck in debt.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Up to now, most IE researchers have been creating paraphrase knowledge (or IE patterns) by hand and for specific tasks.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The second row is the accuracy of the best of the three parsers.'","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"To quantize, we use the binning method (Federico and Bertoldi, 2006) that sorts values, divides into equally sized bins, and averages within each bin.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This problem arises because our keywords consist of only one word.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Instead, we resort to an iterative update based method.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Clearly, explicitly modeling such a powerful constraint on tagging assignment has a potential to significantly improve the accuracy of an unsupervised part-of-speech tagger learned without a tagging dictionary.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"6 Results and Analysis.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Since pronouns carry little semantics of their own, resolving them depends almost entirely on context.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR merely identifies caseframes that frequently co-occur in coreference resolutions.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We have checked if there are similar verbs in other major domains, but this was the only one.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(b) 89 :1 t& tal de cai2neng2 hen3 he DE talent very 'He has great talent' f.b ga ol hig h While the current algorithm correctly handles the (b) sentences, it fails to handle the (a) sentences, since it does not have enough information to know not to group the sequences.ma3lu4 and?]cai2neng2 respectively.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"First, a non-anaphoric NP classifier identifies definite noun phrases that are existential, using both syntactic rules and our learned existential NP recognizer (Bean and Riloff, 1999), and removes them from the resolution process.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"For other languages, we use the CoNLL-X multilingual dependency parsing shared task corpora (Buchholz and Marsi, 2006) which include gold POS tags (used for evaluation).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"MCTAG's are able to generate tee sets having dependent paths.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"However there is no global pruning.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"paper, and is missing 6 examples from the A set.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Trying to integrate constituent ordering and choice of referring expressions, (Chiarcos 2003) developed a numerical model of salience propagation that captures various factors of author\u00e2\u0080\u0099s intentions and of information structure for ordering sentences as well as smaller constituents, and picking appropriate referring expressions.10 Chiarcos used the PCC annotations of co-reference and information structure to compute his numerical models for salience projection across the generated texts.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Feature-based HMM Model (Berg- Kirkpatrick et al., 2010): The KM model uses a variety of orthographic features and employs the EM or LBFGS optimization algorithm; Posterior regulariation model (Grac\u00c2\u00b8a et al., 2009): The G10 model uses the posterior regular- ization approach to ensure tag sparsity constraint.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We resolve this problem by inserting an entry with probability set to an otherwise-invalid value (\u2212oc).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"However, we needed to restrict ourselves to these languages in order to be able to evaluate the performance of our approach.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This fact annoyed especially his dog...).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"+ cost(unseen(fm, as desired.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"0750271 and by the DARPA GALE program.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"But in most cases they can be used interchangably.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Tsarfaty (2006) was the first to demonstrate that fully automatic Hebrew parsing is feasible using the newly available 5000 sentences treebank.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The average individual parser accuracy was reduced by more than 5% when we added this new parser, but the precision of the constituent voting technique was the only result that decreased significantly.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Given the bilingual graph described in the previous section, we can use label propagation to project the English POS labels to the foreign language.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The computation of Pfr1(c)1Mi M k (C)) has been sketched before in Equations 1 through 4.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This model is easily incorporated into the segmenter by building a WFST restrict\u00c2\u00ad ing the names to the four licit types, with costs on the arcs for any particular name summing to an estimate of the cost of that name.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"This enables us to build a high performance NER without using separate classifiers to take care of global consistency or complex formulation on smoothing and backoff models (Bikel et al., 1997).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This decreases the statistical significance of our results compared to those studies.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We currently simulate this by crafting a WCFG and feeding it to BitPar.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Assigning rhetorical relations thus poses questions that can often be answered only subjectively.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"\u00e2\u0080\u009cAgree\u00e2\u0080\u009d is a subject control verb, which dominates another verb whose subject is the same as that of \u00e2\u0080\u009cagree\u00e2\u0080\u009d; the latter verb is generally the one of interest for extraction.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"This is especially true in the case of quotations\u00e2\u0080\u0094which are common in the ATB\u00e2\u0080\u0094where (1) will follow a verb like (2) (Figure 1).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Unsupervised learning approaches appear to be a natural solution to this problem, as they require only unannotated text for training models.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The translation of one position in the source sentence may be postponed for up to L = 3 source positions, and the translation of up to two source positions may be anticipated for at most R = 10 source positions.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"85 82.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If one of these checks fails then this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We reduce this to O(log log |A|) time by evenly distributing keys over their range then using interpolation search4 (Perl et al., 1978).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"So the success of the algorithm may well be due to its success in maximizing the number of unlabeled examples on which the two decision lists agree.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The test accuracy more or less asymptotes.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Although these authors report better gains than ours, they are with respect to a non-adapted baseline.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Translation errors are reported in terms of multireference word error rate (mWER) and subjective sentence error rate (SSER).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Each token may admit multiple analyses, each of which a sequence of one or more lexemes (we use li to denote a lexeme) belonging a presupposed Hebrew lexicon LEX.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"= p(fj je) max \u00c3\u0086;e00 j02Cnfjg np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j 0 )o: The DP equation is evaluated recursively for each hypothesis (e0; e; C; j).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"As mentioned above, it is not obvious how to apply Daum\u00b4e\u2019s approach to multinomials, which do not have a mechanism for combining split features.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This decreases the statistical significance of our results compared to those studies.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The local features used are similar to those used in BBN' s IdentiFinder (Bikel et al., 1999) or MENE (Borthwick, 1999).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The wounds are still healing.), entity-attribute (e.g., She 2001), who determined that in their corpus of German computer tests, 38% of relations were lexically signalled.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Then, moving from connective to connective, ConAno sometimes offers suggestions for its scope (using heuristics like \u00e2\u0080\u0098for sub- junctor, mark all words up to the next comma as the first segment\u00e2\u0080\u0099), which the annotator can accept with a mouseclick or overwrite, marking instead the correct scope with the mouse.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"As far as we know, no other NERs have used information from the whole document (global) as well as information within the same sentence (local) in one framework.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thus, TAG's can not give analyses in which dependencies between arbitrarily large branches exist.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Section 4 draws some conclusions from the present state of the effort.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"On the other hand, if it is seen as McCann Pte.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The details are given in (Och and Ney, 2000).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"By design, they readily capture regularities at the token-level.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Finally, since non-projective constructions often involve long-distance dependencies, the problem is closely related to the recovery of empty categories and non-local dependencies in constituency-based parsing (Johnson, 2002; Dienes and Dubey, 2003; Jijkoun and de Rijke, 2004; Cahill et al., 2004; Levy and Manning, 2004; Campbell, 2004).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"4.3 Morphological Analysis.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Again, the monotone search performs worst.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For the inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J), we drop the dependence on the target sentence length I. 2.2 Word Joining.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"If the same pair of NE instances is used with different phrases, these phrases are likely to be paraphrases.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Jiang and Zhai (2007) suggest the following derivation, making use of the true OUT distribution po(s, t): where each fi(s, t) is a feature intended to charac- !0\u02c6 = argmax pf(s, t) log p\u03b8(s|t) (8) terize the usefulness of (s, t), weighted by Ai. \u03b8 s,t pf(s, t)po(s, t) log p\u03b8(s|t) The mixing parameters and feature weights (col- != argmax po (s, t) lectively 0) are optimized simultaneously using dev- \u03b8 s,t pf(s, t)co(s, t) log p\u03b8(s|t), set maximum likelihood as before: !\ufffdargmax po (s, t) ! \u03b8 s,t \ufffd\u02c6 = argmax \u02dcp(s, t) log p(s|t; 0).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"96 75.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The 14 general-language features embody straightforward cues: frequency, \u201ccentrality\u201d as reflected in model scores, and lack of burstiness.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The work of the second author as well as collaboration visits to Israel was financed by NWO, grant number 017.001.271.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"mein 5.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Clearly this poses a number of research challenges, though, such as the applicability of tag sets across different languages.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We are currently exploring such algorithms.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Chinese word segmentation can be viewed as a stochastic transduction problem.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The algorithm works due to the fact that not all permutations of cities have to be considered explicitly.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Because we are working with only three parsers, the only situation in which context will help us is when it can indicate we should choose to believe a single parser that disagrees with the majority hypothesis instead of the majority hypothesis itself.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Jud ges A G G R ST M 1 M 2 M 3 T1 T2 T3 AG 0.7 0 0.7 0 0 . 4 3 0.4 2 0.6 0 0.6 0 0.6 2 0.5 9 GR 0.9 9 0 . 6 2 0.6 4 0.7 9 0.8 2 0.8 1 0.7 2 ST 0 . 6 4 0.6 7 0.8 0 0.8 4 0.8 2 0.7 4 M1 0.7 7 0.6 9 0.7 1 0.6 9 0.7 0 M2 0.7 2 0.7 3 0.7 1 0.7 0 M3 0.8 9 0.8 7 0.8 0 T1 0.8 8 0.8 2 T2 0.7 8 respectively, the recall and precision.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"There is no global pruning.","target_text":"A modified language model probability p\u00c3\u0086(eje0; e00) is defined as follows: p\u00c3\u0086(eje0; e00) =  1:0 if \u00c3\u0086 = 0 p(eje0; e00) if \u00c3\u0086 = 1 : We associate a distribution p(\u00c3\u0086) with the two cases \u00c3\u0086 = 0 and \u00c3\u0086 = 1 and set p(\u00c3\u0086 = 1) = 0:7.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Evalb is a Java re-implementation of the standard labeled precision\/recall metric.12 The ATB gives all punctuation a single tag.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The work of the second author as well as collaboration visits to Israel was financed by NWO, grant number 017.001.271.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Note that because we extracted only high-confidence alignments, many foreign vertices will not be connected to any English vertices.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"3.2 Stochastic rhetorical analysis.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"By aggregating the POS labels of the English tokens to types, we can generate label distributions for the English vertices.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The contextual role knowledge that BABAR uses for coreference resolution is derived from this caseframe data.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"19 We note that it is not always clear in Wang, Li, and Chang's examples which segmented words.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"More details on the parsing algorithm can be found in Nivre (2003).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In the case of adverbial reduplication illustrated in (3b) an adjective of the form AB is reduplicated as AABB.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"In future work we plan to try this approach with more competitive SMT systems, and to extend instance weighting to other standard SMT components such as the LM, lexical phrase weights, and lexicalized distortion.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"For example, in Information Retrieval (IR), we have to match a user\u00e2\u0080\u0099s query to the expressions in the desired documents, while in Question Answering (QA), we have to find the answer to the user\u00e2\u0080\u0099s question even if the formulation of the answer in the document is different from the question.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As with personal names, we also derive an estimate from text of the probability of finding a transliterated name of any kind (PTN).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"These two restrictions impose the constraint that the result of composing any two structures should be a structure whose &quot;size&quot; is the sum of its constituents plus some constant For example, the operation 4, discussed in the case of CFG's (in Section 4.1) adds the constant equal to the sum of the length of the strings VI, un+r\u2022 Since we are considering formalisms with arbitrary structures it is difficult to precisely specify all of the restrictions on the composition operations that we believe would appropriately generalize the concatenation operation for the particular structures used by the formalism.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"For the error counts, a range from 0:0 to 1:0 is used.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We would also like to thank Amarnag Subramanya for helping us with the implementation of label propagation and Shankar Kumar for access to the parallel data.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"AdaBoost was first introduced in (Freund and Schapire 97); (Schapire and Singer 98) gave a generalization of AdaBoost which we will use in this paper.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Since pronouns carry little semantics of their own, resolving them depends almost entirely on context.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"State is implemented in their scrolling variant, which is a trie annotated with forward and backward pointers.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We have not explored this strategy.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Table 8b shows that verbal nouns are the hardest pre-terminal categories to identify.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"(2010) consistently outperforms ours on English, we obtain substantial gains across other languages.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"An initial step of any text\u00c2\u00ad analysis task is the tokenization of the input into words.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The use of ILP in learning the desired grammar significantly increases the computational complexity of this method.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This information is readily available in TRIE where adjacent records with equal pointers indicate no further extension of context is possible.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"3 These are not full case frames in the traditional sense, but they approximate a simple case frame with a single slot.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The results, along with the total number of phrases, are shown in Table 1.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"J:j:l :zhongl :0.0 ;m,Jlong4 :0.0 (mHHaryg9tltHBI) \u00c2\u00a3: _ADV: 5.98 \u00c2\u00a5 :hua2:o.o E :_NC: 4.41 :mln2:o.o mm : guo2 : 0.0 (RopubllcofChlna) .....,.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In many cases, inspection of either the spelling or context alone is sufficient to classify an example.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We experiment with four values for each hyperparameter resulting in 16 (\u00ce\u00b1, \u00ce\u00b2) combinations: \u00ce\u00b1 \u00ce\u00b2 0.001, 0.01, 0.1, 1.0 0.01, 0.1, 1.0, 10 Iterations In each run, we performed 30 iterations of Gibbs sampling for the type assignment variables W .4 We use the final sample for evaluation.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Learned Tag Prior (PRIOR) We next assume there exists a single prior distribution \u00cf\u0088 over tag assignments drawn from DIRICHLET(\u00ce\u00b2, K ).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Third, we develop a human interpretable grammar that is competitive with a latent variable PCFG.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"32 81.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In this paper, we outline how such family of formalisms can be defined, and show that like CFG's, each member possesses a number of desirable linguistic and computational properties: in particular, the constant growth property and polynomial recognizability.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"annotation guidelines that tell annotators what to do in case of doubt.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"2.1 Part-of-speech tags.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"While there are other obstacles to completing this idea, we believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Next, we describe four contextual role knowledge sources that are created from the training examples and the caseframes.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Before describing the unsupervised case we first describe the supervised version of the algorithm: Input to the learning algorithm: n labeled examples of the form (xi, y\u201e). y, is the label of the ith example (given that there are k possible labels, y, is a member of y = {1 ... 0). xi is a set of mi features {x,1, Xi2 .","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"To explicitly handle the word reordering between words in source and target language, we use the concept of the so-called inverted alignments as given in (Ney et al., 2000).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For each candidate antecedent, BABAR identifies the caseframe that would extract the candidate, pairs it with the anaphor\u00e2\u0080\u0099s caseframe, and consults the CF Network to see if this pair of caseframes has co-occurred in previous resolutions.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Supervised methods have been applied quite successfully to the full MUC named-entity task (Bikel et al. 97).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"(Brandt 1996) extended these ideas toward a conception of kommunikative Gewichtung (\u00e2\u0080\u0098communicative-weight assignment\u00e2\u0080\u0099).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Clearly, for judges h and h taking h as standard and computing the precision and recall for Jz yields the same results as taking h as the standard, and computing for h, 14 All evaluation materials, with the exception of those used for evaluating personal names were drawn.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"An alternate approximation to (8) would be to let w,\\(s, t) directly approximate p\u02c6I(s, t).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The major problem for all segmentation systems remains the coverage afforded by the dictionary and the lexical rules used to augment the dictionary to deal with unseen words.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Also, expanding on the techniques for the automatic generation of extraction patterns (Riloff 96; Sudo 03) using our method, the extraction patterns which have the same meaning can be automatically linked, enabling us to produce the final table fully automatically.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Modifying the Berkeley parser for Arabic is straightforward.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Find keywords for each NE pair The keywords are found for each NE category pair.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"This step ensures that the most frequent terms for each domain are labeled (in case some of them are not in WordNet) and labeled with the sense most appropriate for the domain.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Examples will usually be accompanied by a translation, plus a morpheme-by-morpheme gloss given in parentheses whenever the translation does not adequately serve this purpose.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"A secondary reference resolution classifier has information on the class assigned by the primary classifier.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Full Chinese personal names are in one respect simple: they are always of the form family+given.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The second main result is that the pseudo-projective approach to parsing (using special arc labels to guide an inverse transformation) gives a further improvement of about one percentage point on attachment score.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"But we will show that the use of unlabeled data can drastically reduce the need for supervision.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Making the reasonable assumption that similar information is relevant for solving these problems in Chinese, it follows that a prerequisite for intonation-boundary assignment and prominence assignment is word segmentation.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.1 Part-of-speech tags.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The probability of the classes assigned to the words in a sentence in a document is defined as follows: where is determined by the maximum entropy classifier.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Our implementation permits jumping to any n-gram of any length with a single lookup; this appears to be unique among language model implementations.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Before presenting our results, we describe the datasets that we used, as well as two baselines.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Thus, rather than give a single evaluative score, we prefer to compare the performance of our method with the judgments of several human subjects.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The idea of distinguishing between general and domain-specific examples is due to Daum\u00b4e and Marcu (2006), who used a maximum-entropy model with latent variables to capture the degree of specificity.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"They first collect the NE instance pairs and contexts, just like our method.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The constituent voting and na\u00efve Bayes techniques are equivalent because the parameters learned in the training set did not sufficiently discriminate between the three parsers.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Table 3 contains the results for evaluating our systems on the test set (section 22).","label":0},{"source_text":"Here we present two algorithms.","target_text":"(4) is minimized by setting Since a feature may be present in only a few examples, W_ can be in practice very small or even 0, leading to extreme confidence values.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This is especially bad with PROBING because it is based on hashing and performs random lookups, but it is not intended to be used in low-memory scenarios.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"For \u00c3\u0086 = 0, no new target word is generated, while an additional source sentence position is covered.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We are currently exploring such algorithms.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"By definition, each existential NP uniquely specifies an object or concept, so we can infer that all instances of the same existential NP are coreferent (e.g., \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d always refers to the same entity).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Table 4 shows the results.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Applications The discovered paraphrases have multiple applications.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Each member of a set of trees can be adjoined into distinct nodes of trees in a single elementary tree set, i.e, derivations always involve the adjunction of a derived auxiliary tree set into an elementary tree set.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The corpus has been annotated with six different types of information, which are characterized in the following subsections.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"In this paper, we make a simplifying assumption of one-tag-per-word.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Motivated by these questions, we significantly raise baselines for three existing parsing models through better grammar engineering.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Additionally, for a given coverage set, at most 250 different hypotheses are kept during the search process, and the number of different words to be hypothesized by a source word is limited.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Mohri [1995]) shows promise for improving this situation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Finally, this effort is part of a much larger program that we are undertaking to develop stochastic finite-state methods for text analysis with applications to TIS and other areas; in the final section of this paper we will briefly discuss this larger program so as to situate the work discussed here in a broader context.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In section 2 we introduce the graph transformation techniques used to projectivize and deprojectivize dependency graphs, and in section 3 we describe the data-driven dependency parser that is the core of our system.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"GL is then used to parse the string tn1 ... tnk_1, where tni is a terminal corresponding to the lattice span between node ni and ni+1.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We have shown that, at least given independent human judgments, this is not the case, and that therefore such simplistic measures should be mistrusted.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"While Gan's system incorporates fairly sophisticated models of various linguistic information, it has the drawback that it has only been tested with a very small lexicon (a few hundred words) and on a very small test set (thirty sentences); there is therefore serious concern as to whether the methods that he discusses are scalable.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Let n be some node labeled X in a tree -y (see Figure 3).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The choice between different actions is in general nondeterministic, and the parser relies on a memorybased classifier, trained on treebank data, to predict the next action based on features of the current parser configuration.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Berland and Charniak 99) describe a method for extracting parts of objects from wholes (e.g., &quot;speedometer&quot; from &quot;car&quot;) from a large corpus using hand-crafted patterns.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"computing the recall of the other's judgments relative to this standard.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The NP is a complement to a preposition, which is the head of a PP.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The global feature groups are: InitCaps of Other Occurrences (ICOC): There are 2 features in this group, checking for whether the first occurrence of the same word in an unambiguous position (non first-words in the TXT or TEXT zones) in the same document is initCaps or not-initCaps.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"We can do that . Input: Das ist zu knapp , weil ich ab dem dritten in Kaiserslautern bin . Genaugenommen nur am dritten . Wie ware es denn am ahm Samstag , dem zehnten Februar ? MonS: That is too tight , because I from the third in Kaiserslautern . In fact only on the third . How about ahm Saturday , the tenth of February ? QmS: That is too tight , because I am from the third in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? IbmS: That is too tight , from the third because I will be in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? Input: Wenn Sie dann noch den siebzehnten konnten , ware das toll , ja . MonS: If you then also the seventeenth could , would be the great , yes . QmS: If you could then also the seventeenth , that would be great , yes . IbmS: Then if you could even take seventeenth , that would be great , yes . Input: Ja , das kommt mir sehr gelegen . Machen wir es dann am besten so . MonS: Yes , that suits me perfectly . Do we should best like that . QmS: Yes , that suits me fine . We do it like that then best . IbmS: Yes , that suits me fine . We should best do it like that .","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"An important subproblem of language model storage is therefore sparse mapping: storing values for sparse keys using little memory then retrieving values given keys using little time.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The horizontal axis in this plot represents the most significant dimension, which explains 62% of the variation.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Given names are most commonly two hanzi long, occasionally one hanzi long: there are thus four possible name types, which can be described by a simple set of context-free rewrite rules such as the following: 1.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The English data comes from the WSJ portion of the Penn Treebank and the other languages from the training set of the CoNLL-X multilingual dependency parsing shared task.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The dev and test sets were randomly chosen from the EMEA corpus.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"By definition, each existential NP uniquely specifies an object or concept, so we can infer that all instances of the same existential NP are coreferent (e.g., \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d always refers to the same entity).","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"As long as the main evaluation metric is dependency accuracy per word, with state-of-the-art accuracy mostly below 90%, the penalty for not handling non-projective constructions is almost negligible.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The phrases have to be the expressions of length less than 5 chunks, appear between two NEs.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As we shall argue, the semantic class affiliation of a hanzi constitutes useful information in predicting its properties.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This work was supported in part by the National Science Foundation under grant IRI9704240.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"One conclusion drawn from this annotation effort was that for humans and machines alike, 2 www.sfs.nphil.unituebingen.de\/Elwis\/stts\/ stts.html 3 www.coli.unisb.de\/sfb378\/negra-corpus\/annotate.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"was done by the participants.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"If one is interested in translation, one would probably want to consider show up as a single dictionary word since its semantic interpretation is not trivially derivable from the meanings of show and up.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This is less effective in our setting, where IN and OUT are disparate.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In this paper, we describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For example, suppose one is building a ITS system for Mandarin Chinese.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE (Maximum Entropy Named Entity) (Borth- wick, 1999) was combined with Proteus (a hand- coded system), and came in fourth among all MUC 7 participants.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Other work includes transferring latent topic distributions from source to target language for LM adaptation, (Tam et al., 2007) and adapting features at the sentence level to different categories of sentence (Finch and Sumita, 2008).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Intuitively, it places more weight on OUT when less evidence from IN is available.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Acronyms (ACRO): Words made up of all capitalized letters in the text zone will be stored as acronyms (e.g., IBM).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Notice that the CC-domain is a special case.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The Berkeley parser gives state-of-the-art performance for all metrics.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Even when there is training data available in the domain of interest, there is often additional data from other domains that could in principle be used to improve performance.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"0 57.2 43.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Following Dickinson (2005), we randomly sampled 100 variation nuclei from each corpus and evaluated each sample for the presence of an annotation error.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"For the moment we will assume that there are only two possible labels: each y, is in { \u20141, +1}.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Methods that allow multiple segmentations must provide criteria for choosing the best segmentation.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The ATB is disadvantaged by having fewer trees with longer average 5 LDC A-E catalog numbers: LDC2008E61 (ATBp1v4), LDC2008E62 (ATBp2v3), and LDC2008E22 (ATBp3v3.1).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"There has also been work using a bootstrap- ping approach [Brin 98; Agichtein and Gravano 00; Ravichandran and Hovy 02].","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"\u00e2\u0080\u009d).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"This locally normalized log-linear model can look at various aspects of the observation x, incorporating overlapping features of the observation.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"One obvious application is information extraction.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"besuchen 9.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold, 30.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The first two rows of the table are baselines.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Consider first the examples in (2).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We use a simple TF\/IDF method to measure the topicality of words.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"raphy: A ren2 'person' is a fairly uncontroversial case of a monographemic word, and rplil zhong1guo2 (middle country) 'China' a fairly uncontroversial case of a di\u00c2\u00ad graphernic word.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The sequence of states needed to carry out the word reordering example in Fig.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In this case nonalpha is the string formed by removing all upper\/lower case letters from the spelling (e.g., for Thomas E. Petry nonalpha= .","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"But Arabic contains a variety of linguistic phenomena unseen in English.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"D o m ai n Li n k ac cu ra cy W N c o v e r a g e C C 7 3 . 3 % 2 \/ 1 1 P C 8 8 . 9 % 2 \/ 8 Table 2.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The operations must be linear and nonerasing, i.e., they can not duplicate or erase structure from their arguments.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"And time is short.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"TIS systems in general need to do more than simply compute the.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We train and test on the CoNLL-X training set.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"SRILM\u2019s compact variant, IRSTLM, MITLM, and BerkeleyLM\u2019s sorted variant are all based on this technique.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Supervised part-of-speech (POS) taggers, for example, approach the level of inter-annotator agreement (Shen et al., 2007, 97.3% accuracy for English).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The model we use provides a simple framework in which to incorporate a wide variety of lexical information in a uniform way.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Table 2 shows the features used in the current version of the parser.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"These enable much larger models in memory, compensating for lost accuracy.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Sparse lookup is a key subproblem of language model queries.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We utilized two kinds of datasets in our experiments: (i) monolingual treebanks9 and (ii) large amounts of parallel text with English on one side.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Using structural information As was explained in the results section, we extracted examples like \u00e2\u0080\u009cSmith estimates Lotus\u00e2\u0080\u009d, from a sentence like \u00e2\u0080\u009cMr.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Blum and Mitchell 98) give an example that illustrates just how powerful the second constraint can be.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Cohen and Smith (2007) followed up on these results and proposed a system for joint inference of morphological and syntactic structures using factored models each designed and trained on its own.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For example, it is well-known that one can build a finite-state bigram (word) model by simply assigning a state Si to each word Wi in the vocabulary, and having (word) arcs leaving that state weighted such that for each Wj and corresponding arc aj leaving Si, the cost on aj is the bigram cost of WiWj- (Costs for unseen bigrams in such a scheme would typically be modeled with a special backoff state.)","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Lazy mapping reduces memory requirements by loading pages from disk only as necessary.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Email: gale@research.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"When a token fmnh is to be interpreted as the lexeme sequence f\/REL mnh\/VB, the analysis introduces two distinct entities, the relativizer f (\u201cthat\u201d) and the verb mnh (\u201ccounted\u201d), and not as the complex entity \u201cthat counted\u201d.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In the second and third translation examples, the IbmS word reordering performs worse than the QmS word reordering, since it can not take properly into account the word reordering due to the German verbgroup.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"hanzi in the various name positions, derived from a million names.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"From the point of view of recognition, independent paths in the derivation structures suggests that a top-down parser (for example) can work on each branch independently, which may lead to efficient parsing using an algorithm based on the Divide and Conquer technique.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"paper, and is missing 6 examples from the A set.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"1 2 3.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Next we show that the ATB is similar to other tree- banks in gross statistical terms, but that annotation consistency remains low relative to English (\u00c2\u00a73).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The model was built with open vocabulary, modified Kneser-Ney smoothing, and default pruning settings that remove singletons of order 3 and higher.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Multiple features can be used for the same token.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Previous work has attempted to incorporate such constraints into token-level models via heavy-handed modifications to inference procedure and objective function (e.g., posterior regularization and ILP decoding) (Grac\u00c2\u00b8a et al., 2009; Ravi and Knight, 2009).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"For our experiments we also report the mean of precision and recall, which we denote by (P + R)I2 and F-measure.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Each xii is a member of X, where X is a set of possible features.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Prague Dependency Treebank (Haji\u02c7c et al., 2001b), Danish Dependency Treebank (Kromann, 2003), and the METU Treebank of Turkish (Oflazer et al., 2003), which generally allow annotations with nonprojective dependency structures.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Fourth, we show how to build better models for three different parsers.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"However, the overall percentage of non-projective arcs is less than 2% in PDT and less than 1% in DDT.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The token-level term is similar to the standard HMM sampling equations found in Johnson (2007).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Ex: Mr. Cristiani, president of the country ...","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We have checked if there are similar verbs in other major domains, but this was the only one.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"As we will see from Table 3, not much improvement is derived from this feature.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Finally, we intend to explore more sophisticated instanceweighting features for capturing the degree of generality of phrase pairs.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Only IRSTLM does not support threading.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Here we propose a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The samples from each corpus were independently evaluated.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Pseudo-code describing the generalized boosting algorithm of Schapire and Singer is given in Figure 1.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The points enumerated above are particularly related to ITS, but analogous arguments can easily be given for other applications; see for example Wu and Tseng's (1993) discussion of the role of segmentation in information retrieval.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Figure 1 shows sample sentences from these domains, which are widely divergent.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"tai2du2 'Taiwan Independence.'","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"However, the data sparsity induced by vocalization makes it difficult to train statistical models on corpora of the size of the ATB, so vocalizing and then parsing may well not help performance.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":".., Tn ) T W \u00cf\u0086 E w : Token word seqs (obs) t : Token tag assigns (det by T ) PARAMETERS \u00cf\u0088 : Lexicon parameters \u00ce\u00b8 : Token word emission parameters \u00cf\u0086 : Token tag transition parameters \u00cf\u0086 \u00cf\u0086 t1 t2 \u00ce\u00b8 \u00ce\u00b8 w1 w2 K \u00cf\u0086 T tm O K \u00ce\u00b8 E wN m N N Figure 1: Graphical depiction of our model and summary of latent variables and parameters.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"88,962 (spelling,context) pairs were extracted as training data.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"As indicated by bolding, for seven out of eight languages the improvements of the \u201cWith LP\u201d setting are statistically significant with respect to the other models, including the \u201cNo LP\u201d setting.11 Overall, it performs 10.4% better than the hitherto state-of-the-art feature-HMM baseline, and 4.6% better than direct projection, when we macro-average the accuracy over all languages.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Step 3.","label":0},{"source_text":"There is no global pruning.","target_text":"Final (F): The rest of the sentence is processed monotonically taking account of the already covered positions.","label":0},{"source_text":"There is no global pruning.","target_text":"Covering the first uncovered position in the source sentence, we use the language model probability p(ej$; $).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This is especially effective at reducing load time, since raw bytes are read directly to memory\u2014or, as happens with repeatedly used models, are already in the disk cache.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The approach gains leverage from natural redundancy in the data: for many named-entity instances both the spelling of the name and the context in which it appears are sufficient to determine its type.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As with personal names, we also derive an estimate from text of the probability of finding a transliterated name of any kind (PTN).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"of Articles No.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The first unsupervised algorithm we describe is based on the decision list method from (Yarowsky 95).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"To facilitate comparison with previous work, we exhaustively evaluate this grammar and two other parsing models when gold segmentation is assumed (\u00c2\u00a75).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The segmenter handles the grouping of hanzi into words and outputs word pronunciations, with default pronunciations for hanzi it cannot group; we focus here primarily on the system's ability to segment text appropriately (rather than on its pronunciation abilities).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"A position is presented by the word at that position.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"We show translation results for three approaches: the monotone search (MonS), where no word reordering is allowed (Tillmann, 1997), the quasimonotone search (QmS) as presented in this paper and the IBM style (IbmS) search as described in Section 3.2.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"30 75.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But we follow the more direct adaptation of Evalb suggested by Tsarfaty (2006), who viewed exact segmentation as the ultimate goal.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Both parametric and non-parametric models are explored.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"4 53.7 43.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"So, who won the competition?","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"7 68.3 56.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"A sentence was withheld from section 22 because its extreme length was troublesome for a couple of the parsers.'","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The computing time is given in terms of CPU time per sentence (on a 450MHz PentiumIIIPC).","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"For effectively annotating connectives\/scopes, we found that existing annotation tools were not well-suited, for two reasons: \u00e2\u0080\u00a2 Some tools are dedicated to modes of annotation (e.g., tiers), which could only quite un-intuitively be used for connectives and scopes.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Thus in a two-hanzi word like lflli?J zhong1guo2 (middle country) 'China' there are two syllables, and at the same time two morphemes.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"An examination of the subjects' bracketings confirmed that these instructions were satisfactory in yielding plausible word-sized units.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Assigning rhetorical relations thus poses questions that can often be answered only subjectively.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"However, the overall percentage of non-projective arcs is less than 2% in PDT and less than 1% in DDT.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"As Figure 1 shows, this word has no high-confidence alignment in the Italian-English bitext.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"52 77.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Because the information flow in our graph is asymmetric (from English to the foreign language), we use different types of vertices for each language.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We can now add a new weak hypothesis 14 based on a feature in X1 with a confidence value al hl and atl are chosen to minimize the function We now define, for 1 <i <n, the following virtual distribution, As before, Ztl is a normalization constant.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Here we do not submit to this view.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Benchmarks use the package\u2019s binary format; our code is also the fastest at building a binary file.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"On the one hand, the type-level error rate is not calibrated for the number of n-grams in the sample.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The numbers falling into the location, person, organization categories were 186, 289 and 402 respectively.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For example, in .., says Mr. Cooper, a vice president of.. both a spelling feature (that the string contains Mr.) and a contextual feature (that president modifies the string) are strong indications that Mr. Cooper is of type Person.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"All improvements over the baseline are statistically significant beyond the 0.01 level (McNemar\u2019s test).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The word joining is done on the basis of a likelihood criterion.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Within this framework, we use features intended to capture degree of generality, including the output from an SVM classifier that uses the intersection between IN and OUT as positive examples.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"set was based on an earlier version of the Chang et a!.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Finally, we add \u00e2\u0080\u009cDT\u00e2\u0080\u009d to the tags for definite nouns and adjectives (Kulick et al., 2006).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Conversely, the lattice parser requires no linguistic resources and produces segmentations of comparable quality.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"From now on all lattice arcs are tagged segments and the assignment of probability P(p \u2014* (s, p)) to lattice arcs proceeds as usual.4 A rather pathological case is when our lexical heuristics prune away all segmentation possibilities and we remain with an empty lattice.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"It is difficult to motivate these days why one ministry should be exempt from cutbacks \u00e2\u0080\u0094 at the expense of the others.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In this way we restrict the parameterization of a Language Original case English Danish Dutch German Spanish Swedish Portuguese 94.6 96.3 96.6 95.5 95.4 93.3 95.6 Table 1: Upper bound on tagging accuracy assuming each word type is assigned to majority POS tag.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This is a rather important source of errors in name identifi\u00c2\u00ad cation, and it is not really possible to objectively evaluate a name recognition system without considering the main lexicon with which it is used.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"While we cannot prove there are no such useful features on which one should condition trust, we can give some insight into why the features we explored offered no gain.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Our second contribution is to apply instance weighting at the level of phrase pairs.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In this case, this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We identified three ways that contextual roles can be exploited: (1) by identifying caseframes that co-occur in resolutions, (2) by identifying nouns that co-occur with case- frames and using them to crosscheck anaphor\/candidate compatibility, (3) by identifying semantic classes that co- occur with caseframes and using them to crosscheck anaphor\/candidate compatability.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"0n0'i'i0'2&quot;bin242bn I n = 711 + n2 } On the other hand, no linguistic use is made of this general form of composition and Steedman (personal communication) and Steedman (1986) argues that a more limited definition of composition is more natural.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Note that hanzi that are not grouped into dictionary words (and are not identified as single\u00c2\u00ad hanzi words), or into one of the other categories of words discussed in this paper, are left unattached and tagged as unknown words.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"A list of words occurring more than 10 times in the training data is also collected (commonWords).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The correct resolution in sentence (b) comes from knowledge that people who are kidnapped are often subsequently released.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Then, moving from connective to connective, ConAno sometimes offers suggestions for its scope (using heuristics like \u00e2\u0080\u0098for sub- junctor, mark all words up to the next comma as the first segment\u00e2\u0080\u0099), which the annotator can accept with a mouseclick or overwrite, marking instead the correct scope with the mouse.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Link phrases based on instance pairs Using NE instance pairs as a clue, we find links between sets of phrases.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Unsupervised learning approaches appear to be a natural solution to this problem, as they require only unannotated text for training models.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Assume that the two classifiers are &quot;rote learners&quot;: that is, 1.1 and 12 are defined through look-up tables that list a label for each member of X1 or X2.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"One may argue with these efforts on normalization, and ultimately their value should be assessed by assessing their impact on inter-annotator agreement.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Unification is used in LFG's to produce structures having two dependent spines of unbounded length as in Figure 5.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"However, when the linear LM is combined with a linear TM (lm+lin tm) or MAP TM (lm+map TM), the results are much better than a log-linear combination for the EMEA setting, and on a par for NIST.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The alignment model uses two kinds of parameters: alignment probabilities p(aj jaj\u00f4\u0080\u0080\u00801; I; J), where the probability of alignment aj for position j depends on the previous alignment position aj\u00f4\u0080\u0080\u00801 (Ney et al., 2000) and lexicon probabilities p(fj jeaj ).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Removing the leaves from the resulting tree yields a parse for L under G, with the desired probabilities.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Instead, we extend the variation n-gram method of Dickinson (2005) to compare annotation error rates in the WSJ and ATB.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"There has been additional recent work on inducing lexicons or other knowledge sources from large corpora.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Table 2 shows these similarity measures.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"As discussed in Section 3, independent paths in tree sets, rather than the path complexity, may be crucial in characterizing semilinearity and polynomial time recognition.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"To this end, we construct a bilingual graph over word types to establish a connection between the two languages (\u00a73), and then use graph label propagation to project syntactic information from English to the foreign language (\u00a74).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"We have tested the translation system on the Verbmobil task (Wahlster 1993).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We carried out translation experiments in two different settings.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The above auxiliary quantity satisfies the following recursive DP equation: Qe0 (e; C; j) = Initial Skip Verb Final 1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For example, the Wang, Li, and Chang system fails on the sequence 1:f:p:]nian2 nei4 sa3 in (k) since 1F nian2 is a possible, but rare, family name, which also happens to be written the same as the very common word meaning 'year.'","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"37.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"For the LM, adaptive weights are set as follows: where \u03b1 is a weight vector containing an element \u03b1i for each domain (just IN and OUT in our case), pi are the corresponding domain-specific models, and \u02dcp(w, h) is an empirical distribution from a targetlanguage training corpus\u2014we used the IN dev set for this.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"0 . 8 3 1 0.859 496 76.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"SEGTok(noH) is the segmentation accuracy ignoring mistakes involving the implicit definite article h.11 To evaluate our performance on the tagging task we report CPOS and FPOS corresponding to coarse- and fine-grained PoS tagging results (F1) measure.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Dynamic programming efficiently scores many hypotheses by exploiting the fact that an N-gram language model conditions on at most N \u2212 1 preceding words.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The test accuracy more or less asymptotes.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Pairwise comparison is done using the sign test.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"As can be seen in the example, the first two phrases have a different order of NE names from the last two, so we can determine that the last two phrases represent a reversed relation.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We have shown that the maximum entropy framework is able to use global information directly.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The correct resolution in sentence (b) comes from knowledge that people who are kidnapped are often subsequently released.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"(2010)\u00e2\u0080\u0099s richest model: optimized via either EM or LBFGS, as their relative performance depends on the language.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"However, we do not explore this possibility in the current work.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The size of TRIE is particularly sensitive to F1092 c11, so vocabulary filtering is quite effective at reducing model size.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For each case- frame, BABAR collects the head nouns of noun phrases that were extracted by the caseframe in the training corpus.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Our derivation is slightly different from the one presented in (Schapire and Singer 98) as we restrict at to be positive.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The highestorder N-gram array omits backoff and the index, since these are not applicable.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This FSA I can be segmented into words by composing Id(I) with D*, to form the WFST shown in Figure 2(c), then selecting the best path through this WFST to produce the WFST in Figure 2(d).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"8 1 2.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"This paper is based on work supported in part by DARPA through IBM.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Moses sets the cache size parameter to 50 so we did as well; the resulting cache size is 2.82 GB.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"And then there are decisions that systems typically hard-wire, because the linguistic motivation for making them is not well understood yet.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Smith estimates Lotus will make profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We have not to date explored these various options.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"However, our full model takes advantage of word features not present in Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"For other languages, we use the CoNLL-X multilingual dependency parsing shared task corpora (Buchholz and Marsi, 2006) which include gold POS tags (used for evaluation).","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"All annotations are done with specific tools and in XML; each layer has its own DTD.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"8 1 2.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"In future work we plan to try this approach with more competitive SMT systems, and to extend instance weighting to other standard SMT components such as the LM, lexical phrase weights, and lexicalized distortion.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(Riloff and Shepherd 97) describe a bootstrapping approach for acquiring nouns in particular categories (such as &quot;vehicle&quot; or &quot;weapon&quot; categories).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Although the tag distributions of the foreign words (Eq.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"a classifier.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Figure 1: Translation of PCC sample commentary (STTS)2.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"This might be because our features are more comprehensive than those used by Borthwick.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Again, famous place names will most likely be found in the dictionary, but less well-known names, such as 1PM\u00c2\u00b1 R; bu4lang3-shi4wei2-ke4 'Brunswick' (as in the New Jersey town name 'New Brunswick') will not generally be found.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"As in boosting, the algorithm works in rounds.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"All commentaries have been annotated with rhetorical structure, using RSTTool4 and the definitions of discourse relations provided by Rhetorical Structure Theory (Mann, Thompson 1988).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(3) shows learning curves for CoBoost.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This technique was introduced by Clarkson and Rosenfeld (1997) and is also implemented by IRSTLM and BerkeleyLM\u2019s compressed option.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"There exist a few robust broad-coverage parsers that produce non-projective dependency structures, notably Tapanainen and J\u00a8arvinen (1997) and Wang and Harper (2004) for English, Foth et al. (2004) for German, and Holan (2004) for Czech.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Head Grammars (HG's), introduced by Pollard (1984), is a formalism that manipulates headed strings: i.e., strings, one of whose symbols is distinguished as the head.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This is not ideal for some applications, however.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"( b ) s u p p o r t s c a n d i d a t e i f s e l e c t e d s e m a n t i c t a g s m a t c h t h o s e o f t h e a n a p h o r . Le xic al computes degree of lexical overlap b e t w e e n t h e c a n d i d a t e a n d t h e a n a p h o r . Re cen cy computes the relative distance between the c a n d i d a t e a n d t h e a n a p h o r . Sy nR ole computes relative frequency with which the c a n d i d a t e \u00e2\u0080\u0099 s s y n t a c t i c r o l e o c c u r s i n r e s o l u t i o n s . Figure 4: General Knowledge Sources The Lexical KS returns 1 if the candidate and anaphor are identical, 0.5 if their head nouns match, and 0 otherwise.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In general, different modalities (\u00e2\u0080\u009cplanned to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cagreed to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cbought\u00e2\u0080\u009d) were considered to express the same relationship within an extraction setting.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"2.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In fact, training on the official training data is not suitable as the articles in this data set are entirely about aviation disasters, and the test data is about air vehicle launching.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"4 Evaluation Results.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"There have been other kinds of efforts to discover paraphrase automatically from corpora.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In general, several gross corpus statistics favor the ATB, so other factors must contribute to parsing underperformance.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Thus an explicit assumption about the redundancy of the features \u2014 that either the spelling or context alone should be sufficient to build a classifier \u2014 has been built into the algorithm.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"For displaying and querying the annoated text, we make use of the Annis Linguistic Database developed in our group for a large research effort (\u00e2\u0080\u0098Sonderforschungsbereich\u00e2\u0080\u0099) revolving around 9 2.7 Information structure.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For a language like English, this problem is generally regarded as trivial since words are delimited in English text by whitespace or marks of punctuation.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Previous approaches have tried to find examples that are similar to the target domain.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The second main result is that the pseudo-projective approach to parsing (using special arc labels to guide an inverse transformation) gives a further improvement of about one percentage point on attachment score.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Since guess and gold trees may now have different yields, the question of evaluation is complex.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For all these annotation taks, G\u00c2\u00a8otze developed a series of questions (essentially a decision tree) designed to lead the annotator to the ap propriate judgement.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"AdaBoost.MH can be applied to the problem using these pseudolabels in place of supervised examples.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The choice of the genre commentary resulted from the fact that an investigation of rhetorical structure, its interaction with other aspects of discourse structure, and the prospects for its automatic derivation are the key motivations for building up the corpus.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Previous Work.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Similarly, for all the LCFRS's, discussed in Section 2, we can define the relationship between a structure and the sequence of substrings it spans, and the effect of the composition operations on sequences of substrings.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Figure 4 shows a constituent headed by a process nominal with an embedded adjective phrase.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We call this approach parser switching.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We then extract a set of possible tags tx(y) by eliminating labels whose probability is below a threshold value \u03c4: We describe how we choose \u03c4 in \u00a76.4.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Both (Tsarfaty, 2006; Cohen and Smith, 2007) have shown that a single integrated framework outperforms a completely streamlined implementation, yet neither has shown a single generative model which handles both tasks.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"On the other hand, we are interested in the application of rhetorical analysis or \u00e2\u0080\u0098discourse parsing\u00e2\u0080\u0099 (3.2 and 3.3), in text generation (3.4), and in exploiting the corpus for the development of improved models of discourse structure (3.5).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"This is summarized in Equation 5.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Despite these limitations, a purely finite-state approach to Chinese word segmentation enjoys a number of strong advantages.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"More formally, we start by representing the dictionary D as a Weighted Finite State Trans\u00c2\u00ad ducer (WFST) (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Two annotators received training with the RST definitions and started the process with a first set of 10 texts, the results of which were intensively discussed and revised.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"To control for the effect of the HSPELL-based pruning, we also experimented with a morphological analyzer that does not perform this pruning.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Hyperparameters Our model has two Dirichlet concentration hyperparameters: \u00ce\u00b1 is the shared hyperparameter for the token-level HMM emission and transition distributions.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"While processing the source sentence monotonically, the initial state I is entered whenever there are no uncovered positions to the left of the rightmost covered position.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This best instance-weighting model beats the equivalant model without instance weights by between 0.6 BLEU and 1.8 BLEU, and beats the log-linear baseline by a large margin.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In most cases, however, these expansions come with a steep increase in model complexity, with respect to training procedure and inference time.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In the terrorism domain, 1600 texts were used for training and the 40 test docu X \u00e2\u0088\u00a9Y =\u00e2\u0088\u0085 All sets of hypotheses (and their corresponding belief values) in the current model are crossed with the sets of hypotheses (and belief values) provided by the new evidence.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We evaluate our approach on eight European languages (\u00a76), and show that both our contributions provide consistent and statistically significant improvements.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A minimal requirement for building a Chinese word segmenter is obviously a dictionary; furthermore, as has been argued persuasively by Fung and Wu (1994), one will perform much better at segmenting text by using a dictionary constructed with text of the same genre as the text to be segmented.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The ATB has a much higher fraction of nuclei per tree, and a higher type-level error rate.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"(2009) also report results on English, but on the reduced 17 tag set, which is not comparable to ours).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Assume that the two classifiers are &quot;rote learners&quot;: that is, 1.1 and 12 are defined through look-up tables that list a label for each member of X1 or X2.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This Good\u00c2\u00ad Turing estimate of p(unseen(f,) If,) can then be used in the normal way to define the probability of finding a novel instance of a construction in ir, in a text: p(unseen(f,)) = p(unseen(f,) I f,) p(fn Here p(ir,) is just the probability of any construction in ft as estimated from the frequency of such constructions in the corpus.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For Arabic we M o d e l S y s t e m L e n g t h L e a f A n c e s t o r Co rpu s Sent Exact E v a l b L P LR F1 T a g % B a s e l i n e 7 0 St an for d (v 1.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"f, nan2gual+men0 'pumpkins' is by no means impossible.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Training under this model involves estimation of parameter values for P(y), P(m) and P(x I y).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The sentence length probability p(JjI) is omitted without any loss in performance.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"This overview is illustrated in Figure 1.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We will refer to the semantic classes that co-occur with a caseframe as the semantic expectations of the caseframe.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"One striking example is the error reduction for Spanish, which reduces error by 36.5% and 24.7% for the best and median settings respectively.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The binary language model from Section 5.2 and text phrase table were forced into disk cache before each run.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The hash variant is a reverse trie with hash tables, a more memory-efficient version of SRILM\u2019s default.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Following the setup of Johnson (2007), we use the whole of the Penn Treebank corpus for training and evaluation on English.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The linear LM (lin lm), TM (lin tm) and MAP TM (map tm) used with non-adapted counterparts perform in all cases slightly worse than the log-linear combination, which adapts both LM and TM components.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Backoff-smoothed models estimate this probability based on the observed entry with longest matching history wnf , returning where the probability p(wn|wn\u22121 f ) and backoff penalties b(wn\u22121 i ) are given by an already-estimated model.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In practice, this sparsity constraint is difficult to incorporate in a traditional POS induction system (Me\u00c2\u00b4rialdo, 1994; Johnson, 2007; Gao and Johnson, 2008; Grac\u00c2\u00b8a et al., 2009; Berg-Kirkpatrick et al., 2010).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"If the expression is longer or complicated (like \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d and \u00e2\u0080\u009cA\u00e2\u0080\u0099s purchase of B\u00e2\u0080\u009d), it is called \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d, i.e. a set of phrases which express the same thing or event.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"1 2 3.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Intuitively, as suggested by the example in the introduction, this is the right granularity to capture domain effects.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Adam Pauls provided a pre-release comparison to BerkeleyLM and an initial Java interface.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In Section 5, we try to compare results of MENE, IdentiFinder, and MENERGI.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Past work however, has typically associ n = n P (Ti)P (Wi|Ti) = i=1 1 n K n ated these features with token occurrences, typically in an HMM.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The accuracy of the sets in representing paraphrase ranged from 73% to 99%, depending on the NE categories and set sizes; the accuracy of the links for two evaluated domains was 73% and 86%.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The monotone search performs worst in terms of both error rates mWER and SSER.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"If a phrase does not contain any keywords, the phrase is discarded.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Preprocessing the raw trees improves parsing performance considerably.9 We first discard all trees dominated by X, which indicates errors and non-linguistic text.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Table 4 Differences in performance between our system and Wang, Li, and Chang (1992).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Interpolation search is therefore a form of binary search with better estimates informed by the uniform key distribution.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"5.2 Setup.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Thus at each iteration the method induces at most n x k rules, where k is the number of possible labels (k = 3 in the experiments in this paper). step 3.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Daum\u00b4e (2007) applies a related idea in a simpler way, by splitting features into general and domain-specific versions.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"But it conflates the coordinating and discourse separator functions of wa (<..4.b \u00ef\u00bf\u00bd \u00ef\u00bf\u00bd) into one analysis: conjunction(Table 3).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"5 We choose these two metrics over the Variation Information measure due to the deficiencies discussed in Gao and Johnson (2008).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"This is equivalent to the assumption used in probability estimation for na\u00efve Bayes classifiers, namely that the attribute values are conditionally independent when the target value is given.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"4 7 . 3 8 . 9 2 8 . 8 2 0 . 7 3 2 . 3 3 5 . 2 2 9 . 6 2 7 . 6 1 4 . 2 4 2 . 8 4 5 . 9 4 4 . 3 6 0 . 6 6 1 . 5 4 9 . 9 3 3 . 9 Table 6: Type-level Results: Each cell report the type- level accuracy computed against the most frequent tag of each word type.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Besides information structure, the second main goal is to enhance current models of rhetorical structure.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"To resolve the anaphor, we survey the final belief values assigned to each candidate\u00e2\u0080\u0099s singleton set.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Future extensions of the system might include: 1) An extended translation model, where we use more context to predict a source word.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Cohen and Smith approach this by introducing the \u03b1 hyperparameter, which performs best when optimized independently for each sentence (cf.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, in predicting if a word belongs to a word class, is either true or false, and refers to the surrounding context: if = true, previous word = the otherwise The parameters are estimated by a procedure called Generalized Iterative Scaling (GIS) (Darroch and Ratcliff, 1972).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"For all languages we do not make use of a tagging dictionary.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"In order to evaluate and advance this approach, it helps to feed into the knowledge base data that is already enriched with some of the desired information \u00e2\u0080\u0094 as in PCC.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We did have a problem classifying some modified noun phrases where the modified phrase does not represent a qualified or restricted form of the head, like \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, as these are both represented by the keyword \u00e2\u0080\u009cchairman\u00e2\u0080\u009d.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"This may seem surprising, given the experiments reported in section 4, but the explanation is probably that the non-projective dependencies that can be recovered at all are of the simple kind that only requires a single lift, where the encoding of path information is often redundant.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Then, it can be verified that We can now derive the CoBoost algorithm as a means of minimizing Zco.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"selected; and that recall is defined to be the number of correct hits divided by the number of items that should have been selected.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In addition to the automatic methods, AG, GR, and ST, just discussed, we also added to the plot the values for the current algorithm using only dictionary entries (i.e., no productively derived words or names).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Examples are given in Table 4.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, BABAR learned that agents that \u00e2\u0080\u009cassassinate\u00e2\u0080\u009d or \u00e2\u0080\u009cinvestigate a cause\u00e2\u0080\u009d are usually humans or groups (i.e., organizations).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"76 16.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(a) Of the high frequency phrasal categories, ADJP and SBAR are the hardest to parse.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In this section, we describe how contextual role knowledge is represented and learned.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"i..f,..","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The result of this is shown in Figure 7.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"(1999) did not report using any dictionaries, but mentioned in a footnote that they have added list membership features, which have helped marginally in certain domains.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For novel texts, no lexicon that consists simply of a list of word entries will ever be entirely satisfactory, since the list will inevitably omit many constructions that should be considered words.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For \u00c3\u0086 = 1, a new target language word is generated using the trigram language model p(eje0; e00).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Roughly speaking, a language, L, has the property of semilinearity if the number of occurrences of each symbol in any string is a linear combination of the occurrences of these symbols in some fixed finite set of strings.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"3.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We use a simple TF\/IDF method to measure the topicality of words.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Others depend upon various lexical heuris\u00c2\u00ad tics: for example Chen and Liu (1992) attempt to balance the length of words in a three-word window, favoring segmentations that give approximately equal length for each word.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Each decision determines the inclusion or exclusion of a candidate constituent.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Our system fails in (a) because of$ shenl, a rare family name; the system identifies it as a family name, whereas it should be analyzed as part of the given name.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"All notions of word, with the exception of the orthographic word, are as relevant in Chinese as they are in English, and just as is the case in other languages, a word in Chinese may correspond to one or more symbols in the orthog 1 For a related approach to the problem of word-segrnention in Japanese, see Nagata (1994), inter alia..","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In our model, we associate these features at the type-level in the lexicon.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"5 We choose these two metrics over the Variation Information measure due to the deficiencies discussed in Gao and Johnson (2008).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The index in this array is the vocabulary identifier.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"In Table 1 we see with very few exceptions that the isolated constituent precision is less than 0.5 when we use the constituent label as a feature.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"We can do that . Input: Das ist zu knapp , weil ich ab dem dritten in Kaiserslautern bin . Genaugenommen nur am dritten . Wie ware es denn am ahm Samstag , dem zehnten Februar ? MonS: That is too tight , because I from the third in Kaiserslautern . In fact only on the third . How about ahm Saturday , the tenth of February ? QmS: That is too tight , because I am from the third in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? IbmS: That is too tight , from the third because I will be in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? Input: Wenn Sie dann noch den siebzehnten konnten , ware das toll , ja . MonS: If you then also the seventeenth could , would be the great , yes . QmS: If you could then also the seventeenth , that would be great , yes . IbmS: Then if you could even take seventeenth , that would be great , yes . Input: Ja , das kommt mir sehr gelegen . Machen wir es dann am besten so . MonS: Yes , that suits me perfectly . Do we should best like that . QmS: Yes , that suits me fine . We do it like that then best . IbmS: Yes , that suits me fine . We should best do it like that .","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"4 7 . 3 8 . 9 2 8 . 8 2 0 . 7 3 2 . 3 3 5 . 2 2 9 . 6 2 7 . 6 1 4 . 2 4 2 . 8 4 5 . 9 4 4 . 3 6 0 . 6 6 1 . 5 4 9 . 9 3 3 . 9 Table 6: Type-level Results: Each cell report the type- level accuracy computed against the most frequent tag of each word type.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Model Overview The model starts by generating a tag assignment T for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Given the bilingual graph described in the previous section, we can use label propagation to project the English POS labels to the foreign language.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"3An English sentence with ambiguous PoS assignment can be trivially represented as a lattice similar to our own, where every pair of consecutive nodes correspond to a word, and every possible PoS assignment for this word is a connecting arc.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We focus here instead on adapting the two most important features: the language model (LM), which estimates the probability p(wIh) of a target word w following an ngram h; and the translation models (TM) p(slt) and p(t1s), which give the probability of source phrase s translating to target phrase t, and vice versa.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"This locally normalized log-linear model can look at various aspects of the observation x, incorporating overlapping features of the observation.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"In a model we built with default settings, 1.2% of n + 1-grams were missing their ngram suffix.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Both (Tsarfaty, 2006; Cohen and Smith, 2007) have shown that a single integrated framework outperforms a completely streamlined implementation, yet neither has shown a single generative model which handles both tasks.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"3 68.9 50.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Note that since all English vertices were extracted from the parallel text, we will have an initial label distribution for all vertices in Ve.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We thank members of the MIT NLP group for their suggestions and comments.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The quasi-monotone search performs best in terms of both error rates mWER and SSER.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The commentaries in PCC are all of roughly the same length, ranging from 8 to 10 sentences.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The fraction of buckets that are empty is m\u22121 m , so average lookup time is O( m 1) and, crucially, constant in the number of entries.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"It is generally considered to be fast (Pauls 29 \u2212 1 probabilities and 2' \u2212 2 non-zero backoffs. and Klein, 2011), with a default implementation based on hash tables within each trie node.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Every token is independent of the others, and the sentence lattice is in fact a concatenation of smaller lattices, one for each token.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In German, the verbgroup usually consists of a left and a right verbal brace, whereas in English the words of the verbgroup usually form a sequence of consecutive words.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Like in the co-reference annotation, G\u00c2\u00a8otze\u00e2\u0080\u0099s proposal has been applied by two annotators to the core corpus but it has not been systematically evaluated yet.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For a sequence of hanzi that is a possible name, we wish to assign a probability to that sequence qua name.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"For our experiments we also report the mean of precision and recall, which we denote by (P + R)I2 and F-measure.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Only 2 link in the CC- domain (buy-purchase, acquire-acquisition) and 2 links (trader-dealer and head-chief) in the PC- domain are found in the same synset of Word- Net 2.1 (http:\/\/wordnet.princeton.edu\/).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"First, from a large corpus, we extract all the NE instance pairs.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"splitPUNC restores the convention of the WSJ.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"On the first of these-the B set-our system had 64% recall and 86% precision; on the second-the C set-it had 33% recall and 19% precision.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Of course, we are primarily interested in applying our techniques to languages for which no labeled resources are available.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"3 60.7 50.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Section 2 describes our baseline techniques for SMT adaptation, and section 3 describes the instance-weighting approach.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR uses the log-likelihood statistic (Dunning, 1993) to evaluate the strength of a co-occurrence relationship.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The resulting algorithm has a complexity of O(n!).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Hereafter, each pair of NE categories will be called a domain; e.g. the \u00e2\u0080\u009cCompany \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain, which we will call CC- domain (Step 2).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In application settings, this may be a profitable strategy.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"One of the difficulties in Natural Language Processing is the fact that there are many ways to express the same thing or event.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"For comparison, the completely unsupervised feature-HMM baseline accuracy on the universal POS tags for English is 79.4%, and goes up to 88.7% with a treebank dictionary.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The P (W |T , \u00cf\u0088) term in the lexicon component now decomposes as: n P (W |T , \u00cf\u0088) = n P (Wi|Ti, \u00cf\u0088) i=1 n \u00ef\u00a3\u00ab \u00ef\u00a3\u00b6 tions are not modeled by the standard HMM, which = n \u00ef\u00a3\u00ad n P (v|\u00cf\u0088Ti f )\u00ef\u00a3\u00b8 instead can model token-level frequency.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We use the log-linear tagger of Toutanova et al.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The second weakness is purely conceptual, and probably does not affect the per\u00c2\u00ad formance of the model.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For example, ... fraud related to work on a federally funded sewage plant in Georgia In this case, Georgia is extracted: the NP containing it is a complement to the preposition in; the PP headed by in modifies the NP a federally funded sewage plant, whose head is the singular noun plant.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"2 60.7 56.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"If the semantic expectations do not intersect, then we know that the case- frames extract mutually exclusive types of noun phrases.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We have not explored this strategy.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In the named entity domain these rules were Each of these rules was given a strength of 0.9999.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Most languages that use Roman, Greek, Cyrillic, Armenian, or Semitic scripts, and many that use Indian-derived scripts, mark orthographic word boundaries; however, languages written in a Chinese-derived writ\u00c2\u00ad ing system, including Chinese and Japanese, as well as Indian-derived writing systems of languages like Thai, do not delimit orthographic words.1 Put another way, written Chinese simply lacks orthographic words.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"A simple extension will be used to handle this problem.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"and f,.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"\u00e2\u0080\u0094 similar results have been observed across multiple languages.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Up to now, most IE researchers have been creating paraphrase knowledge (or IE patterns) by hand and for specific tasks.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Among machine learning-based NERs, Identi- Finder has proven to be the best on the official MUC6 and MUC7 test data.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The basic strategy is, for a given pair of entity types, to start with some examples, like several famous book title and author pairs; and find expressions which contains those names; then using the found expressions, find more author and book title pairs.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We have modified Moses (Koehn et al., 2007) to keep our state with hypotheses; to conserve memory, phrases do not keep state.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Clearly this is not the only way to estimate word-frequencies, however, and one could consider applying other methods: in partic\u00c2\u00ad ular since the problem is similar to the problem of assigning part-of-speech tags to an untagged corpus given a lexicon and some initial estimate of the a priori probabilities for the tags, one might consider a more sophisticated approach such as that described in Kupiec (1992); one could also use methods that depend on a small hand-tagged seed corpus, as suggested by one reviewer.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Notice that even though IG's and LFG's involve CFG-like productions, they are (linguistically) fundamentally different from CFG's because the composition operations need not be linear.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Unfortunately, the best completely unsupervised English POS tagger (that does not make use of a tagging dictionary) reaches only 76.1% accuracy (Christodoulopoulos et al., 2010), making its practical usability questionable at best.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Nonetheless, the prospect of a network of annotated discourse resources seems particularly promising if not only a single annotation layer is used but a whole variety of them, so that a systematic search for correlations between them becomes possible, which in turn can lead to more explanatory models of discourse structure.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This distributional sparsity of syntactic tags is not unique to English 1 The source code for the work presented in this paper is available at http:\/\/groups.csail.mit.edu\/rbg\/code\/typetagging\/.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The secondary classifier in (Borthwick, 1999) uses information not just from the current article, but also from the whole test corpus, with an additional feature that indicates if the information comes from the same document or from another document.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"However, the point of RandLM is to scale to even larger data, compensating for this loss in quality.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Because we are interested in applying our techniques to languages for which no labeled resources are available, we paid particular attention to minimize the number of free parameters and used the same hyperparameters for all language pairs.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Applications The discovered paraphrases have multiple applications.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"\u00e2\u0080\u009cAgree\u00e2\u0080\u009d is a subject control verb, which dominates another verb whose subject is the same as that of \u00e2\u0080\u009cagree\u00e2\u0080\u009d; the latter verb is generally the one of interest for extraction.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"P St = n. \u00ce\u00b2 T VARIABLES \u00cf\u0088 Y W : Word types (W1 ,.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Only 2 link in the CC- domain (buy-purchase, acquire-acquisition) and 2 links (trader-dealer and head-chief) in the PC- domain are found in the same synset of Word- Net 2.1 (http:\/\/wordnet.princeton.edu\/).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"(f1; ;mg ; l) 2 (f1; ;mg n fl; l1g ; l0) !","label":0},{"source_text":"Here we present two algorithms.","target_text":"). context=x The context for the entity.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Section 4 presents experimen tal results on two corpora: the MUC4 terrorism corpus, and Reuters texts about natural disasters.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"kann 7.nicht 8.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"2 62.2 39.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In this case, Maury Cooper is extracted.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"What ought to be developed now is an annotation tool that can make use of the format, allow for underspecified annotations and visualize them accordingly.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In order to train this error-correction model, he divided his training corpus into 5 portions of 20% each.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"For brevity we omit the segments from the analysis, and so analysis of the form \u201cfmnh\u201d as f\/REL mnh\/VB is represented simply as REL VB.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"4 70.4 46.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Evaluation of Morphological Analysis.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Our derivation is slightly different from the one presented in (Schapire and Singer 98) as we restrict at to be positive.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In total, for the 2,000 NE category pairs, 5,184 keywords are found.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The authors provided us with a ratio between TPT and SRI under different conditions. aLossy compression with the same weights. bLossy compression with retuned weights. ditions make the value appropriate for estimating repeated run times, such as in parameter tuning.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Dropping the conditioning on 0 for brevity, and letting \u00afc\u03bb(s, t) = c\u03bb(s, t) + yu(s|t), and \u00afc\u03bb(t) = 4Note that the probabilities in (7) need only be evaluated over the support of \u02dcp(s, t), which is quite small when this distribution is derived from a dev set.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"3 61.7 38.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"For developing these mechanisms, the possibility to feed in hand-annotated information is very useful.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"When a company buys another company, a paying event can occur, but these two phrases do not indicate the same event.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Previous work on morphological and syntactic disambiguation in Hebrew used different sets of data, different splits, differing annotation schemes, and different evaluation measures.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We present two algorithms.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"6 Conclusions.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"(f1; ;mg n fl1; l2; l3g ;m) German to English the monotonicity constraint is violated mainly with respect to the German verbgroup.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The next two rows are results of oracle experiments.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Lattice parsing (Chappelier et al., 1999) is an alternative to a pipeline that prevents cascading errors by placing all segmentation options into the parse chart.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"As long as the main evaluation metric is dependency accuracy per word, with state-of-the-art accuracy mostly below 90%, the penalty for not handling non-projective constructions is almost negligible.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Both implementations employ a state object, opaque to the application, that carries information from one query to the next; we discuss both further in Section 4.2.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"An anti-greedy algorithm, AG: instead of the longest match, take the.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"with the number of exactly matching guess trees.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Methods for expanding the dictionary include, of course, morphological rules, rules for segmenting personal names, as well as numeral sequences, expressions for dates, and so forth (Chen and Liu 1992; Wang, Li, and Chang 1992; Chang and Chen 1993; Nie, Jin, and Hannan 1994).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"An analysis of nouns that occur in both the singular and the plural in our database reveals that there is indeed a slight but significant positive correlation-R2 = 0.20, p < 0.005; see Figure 6.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"\"c' 0 + 0 \"0 ' \u00e2\u0080\u00a2 + a n t i g r e e d y x g r e e d y < > c u r r e n t m e t h o d o d i e t . o n l y \u00e2\u0080\u00a2 Taiwan 0 \u00c2\u00b7;; 0 c CD E i5 0\"' 9 9 \u00e2\u0080\u00a2 Mainland \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 -0.30.20.1 0.0 0.1 0.2 Dimension 1 (62%) Figure 7 Classical metric multidimensional scaling of distance matrix, showing the two most significant dimensions.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Performance improvements transfer to the Moses (Koehn et al., 2007), cdec (Dyer et al., 2010), and Joshua (Li et al., 2009) translation systems where our code has been integrated.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Our coreference resolver performed well in two domains, and experiments showed that each contextual role knowledge source contributed valuable information.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If all three sentences are in the same document, then even a human will find it difficult to classify McCann in (1) into either person or organization, unless there is some other information provided.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The natural baseline approach is to concatenate data from IN and OUT.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"There is a (costless) transition between the NC node and f,.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"However, the characterization given in the main body of the text is correct sufficiently often to be useful.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"then define the best segmentation to be the cheapest or best path in Id(I) o D* (i.e., Id(I) composed with the transitive closure of 0).6 Consider the abstract example illustrated in Figure 2.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"com \u00c2\u00a7Cambridge, UK Email: nc201@eng.cam.ac.uk \u00c2\u00a9 1996 Association for Computational Linguistics (a) B ) ( , : & ; ? ' H o w d o y o u s a y o c t o p u s i n J a p a n e s e ? ' (b) P l a u s i b l e S e g m e n t a t i o n I B X I I 1 : & I 0 0 r i 4 w e n 2 z h a n g l y u 2 z e n 3 m e 0 s h u o l ' J a p a n e s e ' ' o c t o p u s ' ' h o w ' ' s a y ' (c) Figure 1 I m p l a u s i b l e S e g m e n t a t i o n [\u00c2\u00a7] lxI 1:&I ri4 wen2 zhangl yu2zen3 me0 shuol 'Japan' 'essay' 'fish' 'how' 'say' A Chinese sentence in (a) illustrating the lack of word boundaries.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The details are given in (Tillmann, 2000).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Daum\u00b4e (2007) applies a related idea in a simpler way, by splitting features into general and domain-specific versions.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The points enumerated above are particularly related to ITS, but analogous arguments can easily be given for other applications; see for example Wu and Tseng's (1993) discussion of the role of segmentation in information retrieval.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"For comparison to information-retrieval inspired baselines, eg (L\u00a8u et al., 2007), we select sentences from OUT using language model perplexities from IN.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"2 60.7 56.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"And time is short.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The weight on each sentence is a value in [0, 1] computed by a perceptron with Boolean features that indicate collection and genre membership.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Consequently, we implemented our own annotation tool ConAno in Java (Stede, Heintze 2004), which provides specifically the functionality needed for our purpose.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Because we are working with only three parsers, the only situation in which context will help us is when it can indicate we should choose to believe a single parser that disagrees with the majority hypothesis instead of the majority hypothesis itself.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"When OUT is large and distinct, its contribution can be controlled by training separate IN and OUT models, and weighting their combination.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"tai2du2 'Taiwan Independence.'","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"conceptual relationship in the discourse.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The second row represents the performance of the median hyperparameter setting.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The two knowledge sources that use semantic expectations, WordSemCFSem and CFSemCFSem, always return values of -1 or 0.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Maamouri et al.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We evaluate our approach on seven languages: English, Danish, Dutch, German, Portuguese, Spanish, and Swedish.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"On the first of these-the B set-our system had 64% recall and 86% precision; on the second-the C set-it had 33% recall and 19% precision.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, the passive voice pattern \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d and the active voice pattern \u00e2\u0080\u009ckidnapped <direct object>\u00e2\u0080\u009d are merged into a single normalized pattern \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d.2 For the sake of sim plicity, we will refer to these normalized extraction patterns as caseframes.3 These caseframes can capture two types of contextual role information: (1) thematic roles corresponding to events (e.g, \u00e2\u0080\u009c<agent> kidnapped\u00e2\u0080\u009d or \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d), and (2) predicate-argument relations associated with both verbs and nouns (e.g., \u00e2\u0080\u009ckidnapped for <np>\u00e2\u0080\u009d or \u00e2\u0080\u009cvehicle with <np>\u00e2\u0080\u009d).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"43 80.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We maintain a separate array for each length n containing all n-gram entries sorted in suffix order.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"1 55.8 38.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"At each point during the derivation, the parser has a choice between pushing the next input token onto the stack \u2013 with or without adding an arc from the token on top of the stack to the token pushed \u2013 and popping a token from the stack \u2013 with or without adding an arc from the next input token to the token popped.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"When a comparison against previous results requires additional pre-processing, we state it explicitly to allow for the reader to replicate the reported results.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Pairwise comparison is done using the sign test.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"A simple lexicalized PCFG with second order Markovization gives relatively poor performance: 75.95% F1 on the test set.8 But this figure is surprisingly competitive with a recent state-of-the-art baseline (Table 7).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The zone to which a token belongs is used as a feature.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The table shows that the lexicon tag frequency predicated by our full model are the closest to the gold standard.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This style of naming is never required and seems to be losing currency.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The ability to redistribute belief values across sets rather than individual hypotheses is key.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"\u2014 I would also like to point out to commissioner Liikanen that it is not easy to take a matter to a national court.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Semilinearity and the closely related constant growth property (a consequence of semilinearity) have been discussed in the context of grammars for natural languages by Joshi (1983\/85) and Berwick and Weinberg (1984).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Preferences for constituent order (especially in languages with relatively free word order) often belong to this group.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"(2009).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Also, the argument has been made that machine translation performance should be evaluated via task-based evaluation metrics, i.e. how much it assists performing a useful task, such as supporting human translators or aiding the analysis of texts.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Hence, for each token, one of the four features zone-TXT, zone- HL, zone-DATELINE, or zone-DD is set to 1, and the other 3 are set to 0.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"These two principles guide experimentation in this framework, and together with the evaluation measures help us decide which specific type of substructure to combine.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Evalb is a Java re-implementation of the standard labeled precision\/recall metric.12 The ATB gives all punctuation a single tag.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In many cases these failures in recall would be fixed by having better estimates of the actual prob\u00c2\u00ad abilities of single-hanzi words, since our estimates are often inflated.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The second algorithm extends ideas from boosting algorithms, designed for supervised learning tasks, to the framework suggested by (Blum and Mitchell 98).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"All experiments use ATB parts 1\u00e2\u0080\u00933 divided according to the canonical split suggested by Chiang et al.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"(2010) reports the best unsupervised results for English.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"In general, the lemma of the previous section does not ensure that all the productions in the combined parse are found in the grammars of the member parsers.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Once HMM parameters (\u00ce\u00b8, \u00cf\u0086) are drawn, a token-level tag and word sequence, (t, w), is generated in the standard HMM fashion: a tag sequence t is generated from \u00cf\u0086.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"IRSTLM (Federico et al., 2008) is an open-source toolkit for building and querying language models.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For example, it is well-known that one can build a finite-state bigram (word) model by simply assigning a state Si to each word Wi in the vocabulary, and having (word) arcs leaving that state weighted such that for each Wj and corresponding arc aj leaving Si, the cost on aj is the bigram cost of WiWj- (Costs for unseen bigrams in such a scheme would typically be modeled with a special backoff state.)","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Not all the layers have been produced for all the texts yet.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"However, supervised methods rely on labeled training data, which is time-consuming and expensive to generate.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In a more recent study than Chang et al., Wang, Li, and Chang (1992) propose a surname-driven, non-stochastic, rule-based system for identifying personal names.17 Wang, Li, and Chang also compare their performance with Chang et al.'s system.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In the second part of the experiment, we applied the inverse transformation based on breadth-first search under the three different encoding schemes.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Since each composition operation is linear and nonerasing, a bounded sequences of substrings associated with the resulting structure is obtained by combining the substrings in each of its arguments using only the concatenation operation, including each substring exactly once.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For example, in predicting if a word belongs to a word class, is either true or false, and refers to the surrounding context: if = true, previous word = the otherwise The parameters are estimated by a procedure called Generalized Iterative Scaling (GIS) (Darroch and Ratcliff, 1972).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"30 16.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Gather phrases using keywords Now, the keyword with the top TF\/ITF score is selected for each phrase.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Other good classes include JADE and GOLD; other bad classes are DEATH and RAT.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In general, different modalities (\u00e2\u0080\u009cplanned to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cagreed to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cbought\u00e2\u0080\u009d) were considered to express the same relationship within an extraction setting.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"This PP modifies another NP, whose head is a singular noun.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"We have tested the translation system on the Verbmobil task (Wahlster 1993).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"First, the training data for the parser is projectivized by applying a minimal number of lifting operations (Kahane et al., 1998) and encoding information about these lifts in arc labels.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We train and test on the CoNLL-X training set.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Though we are not able to calculate their memory usage on our model, results reported in their paper suggest lower memory consumption than TRIE on large-scale models, at the expense of CPU time.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"This paper is based on work supported in part by DARPA through IBM.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Table 6: Example Translations for the Verbmobil task.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"(2008) reported agreement between the teams (measured with Evalb) at 93.8% F1, the level of the CTB.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The algorithm works due to the fact that not all permutations of cities have to be considered explicitly.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For comparison, the completely unsupervised feature-HMM baseline accuracy on the universal POS tags for English is 79.4%, and goes up to 88.7% with a treebank dictionary.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We then gather all phrases with the same keyword.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Because of this threshold, very few NE instance pairs could be used and hence the variety of phrases was also limited.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The number of paths that can be dependent is bounded by the grammar (in fact the maximum cardinality of a tree set determines this bound).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The highestorder N-gram array omits backoff and the index, since these are not applicable.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Each unlabeled pair (x1,i, x2,i) is represented as an edge between nodes corresponding to x1,i and X2,i in the graph.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Although the segmentation requirements for Arabic are not as extreme as those for Chinese, Arabic is written with certain cliticized prepositions, pronouns, and connectives connected to adjacent words.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"ni2ya3 and @5:2 xilya3, respectively.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We also collapse unary chains withidentical basic categories like NP \u00e2\u0086\u0092 NP.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Richer tag sets have been suggested for modeling morphologically complex distinctions (Diab, 2007), but we find that linguistically rich tag sets do not help parsing.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"As noted in Section 1, our code finds the longest matching entry wnf for query p(wn|s(wn\u22121 f ) The probability p(wn|wn\u22121 f ) is stored with wnf and the backoffs are immediately accessible in the provided state s(wn\u22121 When our code walks the data structure to find wnf , it visits wnn, wnn\u22121, ... , wnf .","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Linear probing hash tables must have more buckets than entries, or else an empty bucket will never be found.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The terrorism examples reflect fairly obvious relationships: people who are murdered are killed; agents that \u00e2\u0080\u009creport\u00e2\u0080\u009d things also \u00e2\u0080\u009cadd\u00e2\u0080\u009d and \u00e2\u0080\u009cstate\u00e2\u0080\u009d things; crimes that are \u00e2\u0080\u009cperpetrated\u00e2\u0080\u009d are often later \u00e2\u0080\u009ccondemned\u00e2\u0080\u009d.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"This intuition is born out by the experimental results.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Unsupervised learning approaches appear to be a natural solution to this problem, as they require only unannotated text for training models.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"While Berg-Kirkpatrick et al.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"This is the only important case, because otherwise the simple majority combining technique would pick the correct constituent.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"In future work we plan to try this approach with more competitive SMT systems, and to extend instance weighting to other standard SMT components such as the LM, lexical phrase weights, and lexicalized distortion.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"There are simply not enough votes remaining to allow any of the crossing structures to enter the hypothesized constituent set.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"02 99.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The 8 similarity-to-IN features are based on word frequencies and scores from various models trained on the IN corpus: To avoid numerical problems, each feature was normalized by subtracting its mean and dividing by its standard deviation.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We consider two variants of Berg-Kirkpatrick et al.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The complexity of the algorithm is O(E3 J2 2J), where E is the size of the target language vocabulary.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But gold segmentation is not available in application settings, so a segmenter and parser are arranged in a pipeline.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"wo rd => na m e 2.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"3 60.7 50.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"An easy way to achieve this is to put the domain-specific LMs and TMs into the top-level log-linear model and learn optimal weights with MERT (Och, 2003).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In contrast to the Bayesian HMM, \u00ce\u00b8t is not drawn from a distribution which has support for each of the n word types.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Given counts cn1 where e.g. c1 is the vocabulary size, total memory consumption, in bits, is Our PROBING data structure places all n-grams of the same order into a single giant hash table.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This WFST represents the segmentation of the text into the words AB and CD, word boundaries being marked by arcs mapping between f and part-of-speech labels.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For each case- frame, BABAR collects the head nouns of noun phrases that were extracted by the caseframe in the training corpus.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"(f1; ;mg n fl1; l2g ; l) 4 (f1; ;m \u00f4\u0080\u0080\u0080 1g n fl1; l2; l3g ; l0) !","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We start with noun features since written Arabic contains a very high proportion of NPs.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"This is the only important case, because otherwise the simple majority combining technique would pick the correct constituent.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Both implementations employ a state object, opaque to the application, that carries information from one query to the next; we discuss both further in Section 4.2.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Clearly this poses a number of research challenges, though, such as the applicability of tag sets across different languages.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Finally, a DempsterShafer probabilistic model evaluates the evidence provided by the knowledge sources for all candidate antecedents and makes the final resolution decision.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Final (F): The rest of the sentence is processed monotonically taking account of the already covered positions.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Each feature group can be made up of many binary features.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"In Equations 1 through 3 we develop the model for constructing our parse using na\u00efve Bayes classification.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"0750271 and by the DARPA GALE program.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In (b) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the kidnapping victims, but in (c) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the armed men.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"By considering derivation trees, and thus abstracting away from the details of the composition operation and the structures being manipulated, we are able to state the similarities and differences between the 'This work was partially supported by NSF grants MCS42-19116-CER, MCS82-07294 and DCR-84-10413, ARO grant DAA 29-84-9-0027, and DARPA grant N00014-85-K0018.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Maximizing (7) is thus much faster than a typical MERT run. where co(s, t) are the counts from OUT, as in (6).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"But Arabic contains a variety of linguistic phenomena unseen in English.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"For the experiments, we used four newswire corpora, the Los Angeles Times\/Washington Post, The New York Times, Reuters and the Wall Street Journal, all published in 1995.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Either save money at any cost - or give priority to education.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"0 68.4 48.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Cluster phrases based on Links We now have a set of phrases which share a keyword.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In our coreference resolver, we define \u00ce\u00b8 to be the set of all candidate antecedents for an anaphor.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Giving a recognition algorithm for LCFRL's involves describing the substrings of the input that are spanned by the structures derived by the LCFRS's and how the composition operation combines these substrings.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The second concerns the methods used (if any) to ex\u00c2\u00ad tend the lexicon beyond the static list of entries provided by the machine-readable dictionary upon which it is based.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(Brin 98) ,describes a system for extracting (author, book-title) pairs from the World Wide Web using an approach that bootstraps from an initial seed set of examples.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Given that part-of-speech labels are properties of words rather than morphemes, it follows that one cannot do part-of-speech assignment without having access to word-boundary information.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"example, in Northern Mandarin dialects there is a morpheme -r that attaches mostly to nouns, and which is phonologically incorporated into the syllable to which it attaches: thus men2+r (door+R) 'door' is realized as mer2.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The observed performance gains, coupled with the simplicity of model implementation, makes it a compelling alternative to existing more complex counterparts.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"7 68.3 56.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"A corpus of German newspaper commentaries has been assembled at Potsdam University, and annotated with different linguistic information, to different degrees.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As indicated in Figure 1(c), apart from this correct analysis, there is also the analysis taking B ri4 as a word (e.g., a common abbreviation for Japan), along with X:\u00c2\u00a5 wen2zhangl 'essay\/ and f!!.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This leads to word- and constituent-boundaries discrepancy, which breaks the assumptions underlying current state-of-the-art statistical parsers.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"5 We choose these two metrics over the Variation Information measure due to the deficiencies discussed in Gao and Johnson (2008).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Sima\u2019an et al. (2001) presented parsing results for a DOP tree-gram model using a small data set (500 sentences) and semiautomatic morphological disambiguation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our coreference resolver also incorporates an existential noun phrase recognizer and a DempsterShafer probabilistic model to make resolution decisions.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"This formulation of the constraint feature is equivalent to the use of a tagging dictionary extracted from the graph using a threshold T on the posterior distribution of tags for a given word type (Eq.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"This modification brings the method closer to the DL-CoTrain algorithm described earlier, and is motivated by the intuition that all three labels should be kept healthily populated in the unlabeled examples, preventing one label from dominating \u2014 this deserves more theoretical investigation.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"There is a fairly large body of work on SMT adaptation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"3 Throughout this paper we shall give Chinese examples in traditional orthography, followed.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The following recursive equation is evaluated: Qe0 (e; S; C; j) = (2) = p(fj je) max \u00c3\u0086;e00 np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) max (S0;j0) (S0 ;Cnfjg;j0)!(S;C;j) j02Cnfjg Qe00 (e0; S0;C n fjg; j0)o: The search ends in the hypotheses (I; f1; ; Jg; j).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Finally, a DempsterShafer probabilistic model evaluates the evidence provided by the knowledge sources for all candidate antecedents and makes the final resolution decision.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"8 57.3 +F EA TS be st me dia n 50.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Automatic paraphrase discovery is an important but challenging task.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"4.1 Local Features.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Similarly Figures 1 and 2 show how the isolated constituent precision varies by sentence length and the size of the span of the hypothesized constituent.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Finally, we wish to reiterate an important point.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Memory usage in PROBING is high, though SRILM is even larger, so where memory is of concern we recommend using TRIE, if it fits in memory.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We can now compare this algorithm to that of (Yarowsky 95).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"CHARACTERIZING STRUCTURAL DESCRIPTIONS PRODUCED BY VARIOUS GRAMMATICAL FORMALISMS*","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The first stage identifies a keyword in each phrase and joins phrases with the same keyword into sets.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"This departure from the traditional token-based tagging approach allows us to explicitly capture type- level distributional properties of valid POS tag as signments as part of the model.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The adjunction operation with respect to tree sets (multicomponent adjunction) is defined as follows.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses the log-likelihood statistic (Dunning, 1993) to evaluate the strength of a co-occurrence relationship.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"This locally normalized log-linear model can look at various aspects of the observation x, incorporating overlapping features of the observation.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"(2010)\u00e2\u0080\u0099s richest model: optimized via either EM or LBFGS, as their relative performance depends on the language.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The taggers were trained on datasets labeled with the universal tags.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"By using the output of a hand-coded system such as Proteus, MENE can improve its performance, and can even outperform IdentiFinder (Borthwick, 1999).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Given names are most commonly two hanzi long, occasionally one hanzi long: there are thus four possible name types, which can be described by a simple set of context-free rewrite rules such as the following: 1.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The cities of the traveling salesman problem correspond to source Table 1: DP algorithm for statistical machine translation.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"3 68.4 49.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"(f1; ;mg n fl1g ; l) 3 (f1; ;mg n fl; l1; l2g ; l0) !","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Training under this model involves estimation of parameter values for P(y), P(m) and P(x I y).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The method just described segments dictionary words, but as noted in Section 1, there are several classes of words that should be handled that are not found in a standard dictionary.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Output of the learning algorithm: a function h:Xxy [0, 1] where h(x, y) is an estimate of the conditional probability p(y1x) of seeing label y given that feature x is present.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Hence, trees shown in Figure 8 can not be generated by any MCTAG (but can be generated by an IG) because the number of pairs of dependent paths grows with n. Since the derivation tees of TAG's, MCTAG's, and HG's are local sets, the choice of the structure used at each point in a derivation in these systems does not depend on the context at that point within the derivation.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(8) can now be rewritten5 as which is of the same form as the function Zt used in AdaBoost.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"2 We now introduce a new algorithm for learning from unlabeled examples, which we will call DLCoTrain (DL stands for decision list, the term Cotrain is taken from (Blum and Mitchell 98)).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, there are several reasons why this approach will not in general work: 1.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"It is natural that the larger the data in the domain, the more keywords are found.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In Section 2, we brie y review our approach to statistical machine translation.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The first 3770 trees of the resulting set then were used for training, and the last 418 are used testing.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In (2a), we want to split the two morphemes since the correct analysis is that we have the adverb :1 cai2 'just,' the modal verb neng2 'be able' and the main verb R: Hke4fu2 'overcome'; the competing analysis is, of course, that we have the noun :1 cai2neng2 'talent,' followed by }'lijke4fu2 'overcome.'","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Finally, the statistical method fails to correctly group hanzi in cases where the individual hanzi comprising the name are listed in the dictionary as being relatively high-frequency single-hanzi words.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Oracle results).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Each word is simply tagged with the semantic classes corresponding to all of its senses.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Instead, we want to apply an inverse transformation to recover the underlying (nonprojective) dependency graph.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Also surprising is the low test set OOV rate given the possibility of morphological variation in Arabic.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"4.2 A Sample Segmentation Using Only Dictionary Words Figure 4 shows two possible paths from the lattice of possible analyses of the input sentence B X:\u00c2\u00a5 .:.S:P:l 'How do you say octopus in Japanese?' previously shown in Figure 1.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":1},{"source_text":"All the texts were annotated by two people.","target_text":"While RST (Mann, Thompson 1988) proposed that a single relation hold between adjacent text segments, SDRT (Asher, Lascarides 2003) maintains that multiple relations may hold simultaneously.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The particular classifier used depends upon the noun.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"These words are in turn highly ambiguous, breaking the assumption underlying most parsers that the yield of a tree for a given sentence is known in advance.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A name already mentioned previously in a document may appear in abbreviated form when it is mentioned again later.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"paper, and is missing 6 examples from the A set.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Instead, the designs of the various annotation layers and the actual annotation work are results of a series of diploma theses, of students\u00e2\u0080\u0099 work in course projects, and to some extent of paid assistentships.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"2.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"yields.6 But to its great advantage, it has a high ratio of non-terminals\/terminals (\u00ce\u00bc Constituents \/ \u00ce\u00bc Length).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We now describe the CoBoost algorithm for the named entity problem.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"These two properties of the tree sets are not only linguistically relevant, but also have computational importance.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In fact, training on the official training data is not suitable as the articles in this data set are entirely about aviation disasters, and the test data is about air vehicle launching.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Eight out of the thirteen errors in the high frequency phrases in the CC-domain are the phrases in \u00e2\u0080\u009cagree\u00e2\u0080\u009d.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Section 3 describes the complete coreference resolution model, which uses the contextual role knowledge as well as more traditional coreference features.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Clearly this poses a number of research challenges, though, such as the applicability of tag sets across different languages.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"This modification brings the method closer to the DL-CoTrain algorithm described earlier, and is motivated by the intuition that all three labels should be kept healthily populated in the unlabeled examples, preventing one label from dominating \u2014 this deserves more theoretical investigation.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The cities of the traveling salesman problem correspond to source Table 1: DP algorithm for statistical machine translation.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"123 examples fell into the noise category.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This research was supported in part by the Defense Advanced Research Projects Agency as part of the Translingual Information Detection, Extraction and Summarization (TIDES) program, under Grant N66001001-18917 from the Space and Naval Warfare Systems Center, San Diego, and by the National Science Foundation under Grant IIS00325657.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"2 Chinese ?l* han4zi4 'Chinese character'; this is the same word as Japanese kanji..","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"To resolve the anaphor, we survey the final belief values assigned to each candidate\u00e2\u0080\u0099s singleton set.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Even when there is training data available in the domain of interest, there is often additional data from other domains that could in principle be used to improve performance.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The sources of our dictionaries are listed in Table 2.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"(f1; ;mg n fl1; l2; l3g ;m) German to English the monotonicity constraint is violated mainly with respect to the German verbgroup.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Our coreference resolver performed well in two domains, and experiments showed that each contextual role knowledge source contributed valuable information.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"It also does not prune, so comparing to our pruned model would be unfair.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In Section 3, we introduce our novel concept to word reordering and a DP-based search, which is especially suitable for the translation direction from German to English.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Following the system devised under the Qing emperor Kang Xi, hanzi have traditionally been classified according to a set of approximately 200 semantic radicals; members of a radical class share a particular structural component, and often also share a common meaning (hence the term 'semantic').","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The problem can be represented as a graph with 2N vertices corresponding to the members of X1 and X2.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"BerkeleyLM uses states to optimistically search for longer n-gram matches first and must perform twice as many random accesses to retrieve backoff information.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The relevance of the distinction between, say, phonological words and, say, dictionary words is shown by an example like rpftl_A :;!:Hfllil zhong1hua2 ren2min2 gong4he2-guo2 (China people republic) 'People's Republic of China.'","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Table 4 shows how much the Bayes switching technique uses each of the parsers on the test set.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"There has also been work using a bootstrap- ping approach [Brin 98; Agichtein and Gravano 00; Ravichandran and Hovy 02].","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2009), who also incorporate a sparsity constraint, but does via altering the model objective using posterior regularization.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We used this data to build an unpruned ARPA file with IRSTLM\u2019s improved-kneser-ney option and the default three pieces.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"(Webber et al., 2003)).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We are not claiming that this method is almighty.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Te rro ris m Na tur al Dis ast ers mu rde r of < NP > kill ed <p atie nt > <a ge nt > da ma ged wa s inj ure d in < NP > <a ge nt > rep ort ed <a ge nt > add ed <a ge nt > occ urr ed cau se of < NP > <a ge nt > stat ed <a ge nt > add ed <a ge nt > wr eak ed <a ge nt > cro sse d per pet rat ed <p atie nt > con de mn ed <p atie nt > dri ver of < NP > <a ge nt > car ryi ng Figure 1: Caseframe Network Examples Figure 1 shows examples of caseframes that co-occur in resolutions, both in the terrorism and natural disaster domains.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For each terminal, the Leaf Ancestor metric extracts the shortest path to the root.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We run the baseline Moses system for the French-English track of the 2011 Workshop on Machine Translation,9 translating the 3003-sentence test set.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The maximum precision row is the upper bound on accuracy if we could pick exactly the correct constituents from among the constituents suggested by the three parsers.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Among the IS-units, the referring expressions are marked as such and will in the second phase receive a label for cognitive status (active, accessible- text, accessible-situation, inferrable, inactive).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Therefore, we want state to encode the minimum amount of information necessary to properly compute language model scores, so that the decoder will be faster and make fewer search errors.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"(2010), we adopt a simpler na\u00c2\u00a8\u00c4\u00b1ve Bayes strategy, where all features are emitted independently.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"(2) Description Source Location Names http:\/\/www.timeanddate.com http:\/\/www.cityguide.travel-guides.com http:\/\/www.worldtravelguide.net Corporate Names http:\/\/www.fmlx.com Person First Names http:\/\/www.census.gov\/genealogy\/names Person Last Names Table 2: Sources of Dictionaries The McCann family . . .","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Several papers report the use of part-of-speech information to rank segmentations (Lin, Chiang, and Su 1993; Peng and Chang 1993; Chang and Chen 1993); typically, the probability of a segmentation is multiplied by the probability of the tagging(s) for that segmentation to yield an estimate of the total probability for the analysis.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"However, the data sparsity induced by vocalization makes it difficult to train statistical models on corpora of the size of the ATB, so vocalizing and then parsing may well not help performance.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Although this is not a precise criterion, most cases we evaluated were relatively clear-cut.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Finally, we model the probability of a new transliterated name as the product of PTN and PTN(hanzi;) for each hanzi; in the putative name.13 The foreign name model is implemented as an WFST, which is then summed with the WFST implementing the dictionary, morpho 13 The current model is too simplistic in several respects.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"pronunciations of individual words; they also need to compute intonational phrase boundaries in long utterances and assign relative prominence to words in those utterances.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Equ.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"First of all, most previous articles report perfor\u00c2\u00ad mance in terms of a single percent-correct score, or else in terms of the paired measures of precision and recall.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"With each iteration more examples are assigned labels by both classifiers, while a high level of agreement (> 94%) is maintained between them.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For effectively annotating connectives\/scopes, we found that existing annotation tools were not well-suited, for two reasons: \u00e2\u0080\u00a2 Some tools are dedicated to modes of annotation (e.g., tiers), which could only quite un-intuitively be used for connectives and scopes.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"There have been other kinds of efforts to discover paraphrase automatically from corpora.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"This left 962 examples, of which 85 were noise.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"It would have therefore also been possible to use the integer programming (IP) based approach of Ravi and Knight (2009) instead of the feature-HMM for POS induction on the foreign side.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We consider two variants of Berg-Kirkpatrick et al.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"N, portion of examples on which both classifiers give a label rather than abstaining), and the proportion of these examples on which the two classifiers agree.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Exploiting Diversity in Natural Language Processing: Combining Parsers","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In (2a), we want to split the two morphemes since the correct analysis is that we have the adverb :1 cai2 'just,' the modal verb neng2 'be able' and the main verb R: Hke4fu2 'overcome'; the competing analysis is, of course, that we have the noun :1 cai2neng2 'talent,' followed by }'lijke4fu2 'overcome.'","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The availability of comparable corpora is limited, which is a significant limitation on the approach.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"If 7 is an elementary tree, the derivation tree consists of a single node labeled 7.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The pseudo-code describing the algorithm is given in Fig.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"4 To be sure, it is not always true that a hanzi represents a syllable or that it represents a morpheme.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"This is the first set that gives us a fair evaluation of the Bayes models, and the Bayes switching model performs significantly better than its non-parametric counterpart.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"While Berg-Kirkpatrick et al.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Backoff-smoothed models estimate this probability based on the observed entry with longest matching history wnf , returning where the probability p(wn|wn\u22121 f ) and backoff penalties b(wn\u22121 i ) are given by an already-estimated model.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Feature weights were set using Och\u2019s MERT algorithm (Och, 2003).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The only supervision is in the form of 7 seed rules (namely, that New York, California and U.S. are locations; that any name containing Mr is a person; that any name containing Incorporated is an organization; and that I.B.M. and Microsoft are organizations).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"A Brief Introduction to the Chinese Writing System Most readers will undoubtedly be at least somewhat familiar with the nature of the Chinese writing system, but there are enough common misunderstandings that it is as well to spend a few paragraphs on properties of the Chinese script that will be relevant to topics discussed in this paper.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Somewhat surprisingly, there do not appear to be large systematic differences between linear and MAP combinations.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"For verbs we add two features.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Sentence (2) and (3) help to disambiguate one way or the other.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"As the name implies, space is O(m) and linear in the number of entries.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"We have tested the translation system on the Verbmobil task (Wahlster 1993).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"JI!","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We do not attempt to identify the types of relationships that are found.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The best analysis of the corpus is taken to be the true analysis, the frequencies are re-estimated, and the algorithm is repeated until it converges.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Acknowledgments We thank Steven Bethard, Evan Rosen, and Karen Shiells for material contributions to this work.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Various verbal (e.g., \u00ef\u00bf\u00bd, .::) and adjectival.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"immediately by a Romanization into the pinyin transliteration scheme; numerals following each pinyin syllable represent tones.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Benchmarks use the package\u2019s binary format; our code is also the fastest at building a binary file.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Since MUC6, BBN' s Hidden Markov Model (HMM) based IdentiFinder (Bikel et al., 1997) has achieved remarkably good performance.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"(2009) on Portuguese (Grac\u00c2\u00b8a et al.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"All the sentences have been analyzed by our chunker and NE tag- ger.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Supervised learning approaches have advanced the state-of-the-art on a variety of tasks in natural language processing, resulting in highly accurate systems.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The cost is computed as follows, where N is the corpus size and f is the frequency: (1) Besides actual words from the base dictionary, the lexicon contains all hanzi in the Big 5 Chinese code\/ with their pronunciation(s), plus entries for other characters that can be found in Chinese text, such as Roman letters, numerals, and special symbols.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"While Berg-Kirkpatrick et al.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Fourth, we show how to build better models for three different parsers.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Ignoring the identity of the target language words e and e0, the possible partial hypothesis extensions due to the IBM restrictions are shown in Table 2.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"0 55.3 34.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This larger corpus was kindly provided to us by United Informatics Inc., R.O.C. a set of initial estimates of the word frequencies.9 In this re-estimation procedure only the entries in the base dictionary were used: in other words, derived words not in the base dictionary and personal and foreign names were not used.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The transitive closure of the dictionary in (a) is composed with Id(input) (b) to form the WFST (c).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Note that the backoff model assumes that there is a positive correlation between the frequency of a singular noun and its plural.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Nonetheless, parse quality is much lower in the joint model because a lattice is effectively a long sentence.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Fortunately, there are only a few hundred hanzi that are particularly common in transliterations; indeed, the commonest ones, such as E. bal, m er3, and iij al are often clear indicators that a sequence of hanzi containing them is foreign: even a name like !:i*m xia4mi3-er3 'Shamir,' which is a legal Chi\u00c2\u00ad nese personal name, retains a foreign flavor because of liM.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In figure 4, reverse relations are indicated by `*\u00e2\u0080\u0099 next to the frequency.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"If it starts with a lower case letter, and contains both upper and lower case letters, then (mixedCaps, zone) is set to 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"To formalize the approach, we introduce four verbgroup states S: Initial (I): A contiguous, initial block of source positions is covered.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"By design, they readily capture regularities at the token-level.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We have shown that, at least given independent human judgments, this is not the case, and that therefore such simplistic measures should be mistrusted.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"2.1 Overview.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Not surprisingly some semantic classes are better for names than others: in our corpora, many names are picked from the GRASS class but very few from the SICKNESS class.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Less frequently studied is the interplay among language, annotation choices, and parsing model design (Levy and Manning, 2003; Ku\u00c2\u00a8 bler, 2005).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Multiple features can be used for the same token.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"30 75.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"This is because in many cases, the use of hyphens can be considered to be optional (e.g., third-quarter or third quarter).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We then gather all phrases with the same keyword.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"37.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Specifically, we (+FEATS) utilizes the tag prior as well as features (e.g., suffixes and orthographic features), discussed in Section 3, for the P (W |T , \u00cf\u0088) component.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"So, we set a threshold that at least two examples are required to build a link.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In order to handle the necessary word reordering as an optimization problem within our dynamic programming approach, we describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming (Held, Karp, 1962).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"(2009).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Various verbal (e.g., \u00ef\u00bf\u00bd, .::) and adjectival.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In any event, to date, we have not compared different methods for deriving the set of initial frequency estimates.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In the pinyin transliterations a dash(-) separates syllables that may be considered part of the same phonological word; spaces are used to separate plausible phonological words; and a plus sign (+) is used, where relevant, to indicate morpheme boundaries of interest.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Here, all tokeninternal collocations of tags unseen in our training data are pruned away.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The model was parameterized such that the joint probability of a (label, feature-set) pair P(yi, xi) is written as The model assumes that (y, x) pairs are generated by an underlying process where the label is first chosen with some prior probability P(yi); the number of features mi is then chosen with some probability P(mi); finally the features are independently generated with probabilities P(xulyi).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Second, rather than relying on a division of the corpus into manually-assigned portions, we use features intended to capture the usefulness of each phrase pair.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We used these three parsers to explore parser combination techniques.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For MUC6, for example, Corporate- Suffix-List is made up of ltd., associates, inc., co, corp, ltd, inc, committee, institute, commission, university, plc, airlines, co., corp. and Person-Prefix- List is made up of succeeding, mr., rep., mrs., secretary, sen., says, minister, dr., chairman, ms. . For a token that is in a consecutive sequence of init then a feature Corporate-Suffix is set to 1.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Let us consider an example of ambiguity caused by devocalization.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The human evaluators were a non-native, fluent Arabic speaker (the first author) for the ATB and a native English speaker for the WSJ.7 Table 5 shows type- and token-level error rates for each corpus.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The data structure was populated with 64-bit integers sampled uniformly without replacement.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"and \u00e2\u0080\u009cH\u00e2\u0080\u009d represents \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"example, in Northern Mandarin dialects there is a morpheme -r that attaches mostly to nouns, and which is phonologically incorporated into the syllable to which it attaches: thus men2+r (door+R) 'door' is realized as mer2.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Statistics for all data sets are shown in Table 2.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Because there might be some controversy about the exact definitions of such universals, this set of coarse-grained POS categories is defined operationally, by collapsing language (or treebank) specific distinctions to a set of categories that exists across all languages.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"As the name implies, space is O(m) and linear in the number of entries.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The prediction based on these features is a knearest neighbor classification, using the IB1 algorithm and k = 5, the modified value difference metric (MVDM) and class voting with inverse distance weighting, as implemented in the TiMBL software package (Daelemans et al., 2003).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This suggests a direct parallel to (1): where \u02dcp(s, t) is a joint empirical distribution extracted from the IN dev set using the standard procedure.2 An alternative form of linear combination is a maximum a posteriori (MAP) combination (Bacchiani et al., 2004).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"For the error counts, a range from 0:0 to 1:0 is used.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Here we use a slightly different notion of lift, applying to individual arcs and moving their head upwards one step at a time: Intuitively, lifting an arc makes the word wk dependent on the head wi of its original head wj (which is unique in a well-formed dependency graph), unless wj is a root in which case the operation is undefined (but then wj \u2014* wk is necessarily projective if the dependency graph is well-formed).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"This distributional sparsity of syntactic tags is not unique to English 1 The source code for the work presented in this paper is available at http:\/\/groups.csail.mit.edu\/rbg\/code\/typetagging\/.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"(If the TF\/IDF score of that word is below a threshold, the phrase is discarded.)","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Linguistic intuitions like those in the previous section inform language-specific annotation choices.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"7 Unlike Dickinson (2005), we strip traces and only con-.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Thus it is possible, for illustration, to look for a noun phrase (syntax tier) marked as topic (information structure tier) that is in a bridging relation (co-reference tier) to some other noun phrase.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The resulting algorithm is depicted in Table 1.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The developers suggested some changes, such as building the model from scratch with IRSTLM, but these did not resolve the problem.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"am 11.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"3.1 Corpora.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We use label propagation in two stages to generate soft labels on all the vertices in the graph.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"It differs from previous machine learning-based NERs in that it uses information from the whole document to classify each word, with just one classifier.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The resulting structural differences between tree- banks can account for relative differences in parsing performance.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"So, we set a threshold that at least two examples are required to build a link.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"All the links in the \u00e2\u0080\u009cCC-domain are shown in Step 4 in subsection 3.2.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The role that each noun phrase plays in the kidnapping event is key to distinguishing these cases.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In Section 4, we present the performance measures used and give translation results on the Verbmobil task.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The translation scores for the hypotheses generated with different threshold values t0 are compared to the translation scores obtained with a conservatively large threshold t0 = 10:0 . For each test series, we count the number of sentences whose score is worse than the corresponding score of the test series with the conservatively large threshold t0 = 10:0, and this number is reported as the number of search errors.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We paid particular attention to minimize the number of free parameters, and used the same hyperparameters for all language pairs, rather than attempting language-specific tuning.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In this paper we have argued that Chinese word segmentation can be modeled ef\u00c2\u00ad fectively using weighted finite-state transducers.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In general, m; l; l0 6= fl1; l2; l3g and in line umber 3 and 4, l0 must be chosen not to violate the above reordering restriction.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Asking the annotator to also formulate the question is a way of arriving at more reproducible decisions.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We applied the AutoSlog system (Riloff, 1996) to our unannotated training texts to generate a set of extraction patterns for each domain.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"8 57.3 +F EA TS be st me dia n 50.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Borth- wick (1999) successfully made use of other hand- coded systems as input for his MENE system, and achieved excellent results.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The P (W |T , \u00cf\u0088) term in the lexicon component now decomposes as: n P (W |T , \u00cf\u0088) = n P (Wi|Ti, \u00cf\u0088) i=1 n \u00ef\u00a3\u00ab \u00ef\u00a3\u00b6 tions are not modeled by the standard HMM, which = n \u00ef\u00a3\u00ad n P (v|\u00cf\u0088Ti f )\u00ef\u00a3\u00b8 instead can model token-level frequency.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The wounds are still healing.), entity-attribute (e.g., She 2001), who determined that in their corpus of German computer tests, 38% of relations were lexically signalled.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A multicomponent Tree Adjoining Grammar (MCTAG) consists of a finite set of finite elementary tree sets.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We resolve this problem by inserting an entry with probability set to an otherwise-invalid value (\u2212oc).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The type-level posterior term can be computed according to, P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2) \u00e2\u0088\u009d Note that each round of sampling Ti variables takes time proportional to the size of the corpus, as with the standard token-level HMM.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"As expected, the most informative encoding, Head+Path, gives the highest accuracy with over 99% of all non-projective arcs being recovered correctly in both data sets.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Recall.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This left 962 examples, of which 85 were noise.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In the second scheme, Head+Path, we in addition modify the label of every arc along the lifting path from the syntactic to the linear head so that if the original label is p the new label is p\u2193.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"It is well-known that constituency parsing models designed for English often do not generalize easily to other languages and treebanks.1 Explanations for this phenomenon have included the relative informativeness of lexicalization (Dubey and Keller, 2003; Arun and Keller, 2005), insensitivity to morphology (Cowan and Collins, 2005; Tsarfaty and Sima\u00e2\u0080\u0099an, 2008), and the effect of variable word order (Collins et al., 1999).","label":1},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"They first collect the NE instance pairs and contexts, just like our method.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The evidence may not say anything about whether A is more likely than B, only that C and D are not likely.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Our evaluation includes both weighted and un- weighted lattices.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A moment's reflection will reveal that things are not quite that simple.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Projectivizing a dependency graph by lifting nonprojective arcs is a nondeterministic operation in the general case.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If a candidate has a belief value \u00e2\u0089\u00a5 .50, then we select that candidate as the antecedent for the anaphor.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Following Dickinson (2005), we randomly sampled 100 variation nuclei from each corpus and evaluated each sample for the presence of an annotation error.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Both (Tsarfaty, 2006; Cohen and Smith, 2007) have shown that a single integrated framework outperforms a completely streamlined implementation, yet neither has shown a single generative model which handles both tasks.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We define an ATM, M, recognizing a language generated by a grammar, G, having the properties discussed in Section 43.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The segmentation chosen is the best path through the WFST, shown in (d).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Our test machine has two Intel Xeon E5410 processors totaling eight cores, 32 GB RAM, and four Seagate Barracuda disks in software RAID 0 running Linux 2.6.18.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Also, \u00e2\u0080\u009cagree\u00e2\u0080\u009d in the CC-domain is not a desirable keyword.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We refer to this process as Reliable Case Resolution because it involves finding cases of anaphora that can be easily resolved with their antecedents.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"A list of words occurring more than 10 times in the training data is also collected (commonWords).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The discovered paraphrases can be a big help to reduce human labor and create a more comprehensive pattern set.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Although the ultimate goal of the Verbmobil project is the translation of spoken language, the input used for the translation experiments reported on in this paper is the (more or less) correct orthographic transcription of the spoken sentences.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This is because in many cases, the use of hyphens can be considered to be optional (e.g., third-quarter or third quarter).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"In each figure the upper graph shows the isolated constituent precision and the bottom graph shows the corresponding number of hypothesized constituents.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"What ought to be developed now is an annotation tool that can make use of the format, allow for underspecified annotations and visualize them accordingly.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The local features used are similar to those used in BBN' s IdentiFinder (Bikel et al., 1999) or MENE (Borthwick, 1999).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"How do additional ambiguities caused by devocalization affect statistical learning?","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"It has no syntactic function.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Thus, rather than give a single evaluative score, we prefer to compare the performance of our method with the judgments of several human subjects.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The tool we use is MMAX8, which has been specifically designed for marking co-reference.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Each pattern represents the role that a noun phrase plays in the surrounding context.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Section 2.2 then describes our representation for contextual roles and four types of contextual role knowledge that are learned from the training examples.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We elected run Moses single-threaded to minimize the impact of RandLM\u2019s cache on memory use.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"(Kehler, 1997) also used a DempsterShafer model to merge evidence from different sources for template-level coreference.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"See Table 2 for the tag set size of other languages.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each case- frame, BABAR collects the head nouns of noun phrases that were extracted by the caseframe in the training corpus.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"For each experiment we gave an nonparametric and a parametric technique for combining parsers.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"BABAR uses a DempsterShafer decision model (Stefik, 1995) to combine the evidence provided by the knowledge sources.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For example Chen and Liu (1992) report precision and recall rates of over 99%, but this counts only the words that occur in the test corpus that also occur in their dictionary.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"However, as we have noted, nothing inherent in the approach precludes incorporating higher-order constraints, provided they can be effectively modeled within a finite-state framework.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In this paper, we will propose an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Second, comparisons of different methods are not meaningful unless one can eval\u00c2\u00ad uate them on the same corpus.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Since the segmentation corresponds to the sequence of words that has the lowest summed unigram cost, the segmenter under discussion here is a zeroth-order model.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Finally, our U (unparsed) measure is used to report the number of sentences to which our system could not propose a joint analysis.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"However, this optimistic search would not visit the entries necessary to store backoff information in the outgoing state.","label":0},{"source_text":"Here we present two algorithms.","target_text":"For a description of the application of AdaBoost to various NLP problems see the paper by Abney, Schapire, and Singer in this volume.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Therefore in cases where the segmentation is identical between the two systems we assume that tagging is also identical.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The normalization factor plays an important role in the AdaBoost algorithm.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"For English POS tagging, BergKirkpatrick et al. (2010) found that this direct gradient method performed better (>7% absolute accuracy) than using a feature-enhanced modification of the Expectation-Maximization (EM) algorithm (Dempster et al., 1977).8 Moreover, this route of optimization outperformed a vanilla HMM trained with EM by 12%.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Although this is not a precise criterion, most cases we evaluated were relatively clear-cut.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Language models that contain wi must also contain prefixes wi for 1 G i G k. Therefore, when the model is queried for p(wnjwn\u22121 1 ) but the longest matching suffix is wnf , it may return state s(wn1) = wnf since no longer context will be found.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"was done by the participants.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The machine learning community has been in a similar situation and has studied the combination of multiple classifiers (Wolpert, 1992; Heath et al., 1996).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Hence, there is a total of 29 classes (7 name classes 4 sub-classes 1 not-a-name class).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Although matching is done at the sentence level, this information is subsequently discarded when all matches are pooled.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Since Dane\u00cb\u0087s\u00e2\u0080\u0099 proposals of \u00e2\u0080\u0098thematic development patterns\u00e2\u0080\u0099, a few suggestions have been made as to the existence of a level of discourse structure that would predict the information structure of sentences within texts.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"52 77.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Each round is composed of two stages; each stage updates one of the classifiers while keeping the other classifier fixed.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Maamouri et al.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Domain adaptation is a common concern when optimizing empirical NLP applications.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We used the TF\/ITF metric to identify keywords.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"When a company buys another company, a paying event can occur, but these two phrases do not indicate the same event.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In this subsection, we will report the results of the experiment, in terms of the number of words, phrases or clusters.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Since the lattice L for a given sentence W is determined by the morphological analyzer M we have which is precisely the formula corresponding to the so-called lattice parsing familiar from speech recognition.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The system of Berg-Kirkpatrick et al.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"60 ( P e tr o v, 2 0 0 9 ) all B e r k e l e y ( S e p . 0 9 ) B a s e l i n e 7 0 a l l G o l d P O S 70 \u00e2\u0080\u0094 \u00e2\u0080\u0094 \u00e2\u0080\u0094 0 . 8 0 9 0.839 335 0 . 7 9","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In Section 5, we try to compare results of MENE, IdentiFinder, and MENERGI.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"There are 13 types of extensions needed to describe the verbgroup reordering.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Sie.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This flexibility, along with the simplicity of implementation and expansion, makes this framework an attractive base for continued research.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"It is important to bear in mind, though, that this is not an inherent limitation of the model.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"La ng ua ge 1T W + P RI O R + F E A T S E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 2 1.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"We call this pseudoprojective dependency parsing, since it is based on a notion of pseudo-projectivity (Kahane et al., 1998).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We experiment with four values for each hyperparameter resulting in 16 (\u00ce\u00b1, \u00ce\u00b2) combinations: \u00ce\u00b1 \u00ce\u00b2 0.001, 0.01, 0.1, 1.0 0.01, 0.1, 1.0, 10 Iterations In each run, we performed 30 iterations of Gibbs sampling for the type assignment variables W .4 We use the final sample for evaluation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Again, this deserves further investigation.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"As was explained in the results section, \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d are not desirable keywords in the CC-domain.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"\u00e2\u0080\u00a2 Bridging links: the annotator is asked to specify the type as part-whole, cause-effect (e.g., She had an accident.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"shortest match at each point.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This latter evaluation compares the performance of the system with that of several human judges since, as we shall show, even people do not agree on a single correct way to segment a text.","label":1},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"In this way we restrict the parameterization of a Language Original case English Danish Dutch German Spanish Swedish Portuguese 94.6 96.3 96.6 95.5 95.4 93.3 95.6 Table 1: Upper bound on tagging accuracy assuming each word type is assigned to majority POS tag.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"i=1 (f,v)\u00e2\u0088\u0088Wi","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"How do additional ambiguities caused by devocalization affect statistical learning?","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"(f1; ;mg n fl1; l2; l3g ;m) German to English the monotonicity constraint is violated mainly with respect to the German verbgroup.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"An example is in (i), where the system fails to group t;,f;?\"$?t!: lin2yang2gang3 as a name, because all three hanzi can in principle be separate words (t;,f; lin2 'wood';?\"$ yang2 'ocean'; ?t!; gang3 'harbor').","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Limitations There are several limitations in the methods.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A Person-Prefix-List is compiled in an analogous way.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Our new approach, which is called quasi-monotone search, processes the source sentence monotonically, while explicitly taking into account the positions of the German verbgroup.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Lexicalizing several POS tags improves performance.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The translation direction is from German to English.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The distinctions in the ATB are linguistically justified, but complicate parsing.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The F- measure score increased for both domains, reflecting a substantial increase in recall with a small decrease in precision.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The foreign language vertices (denoted by Vf) correspond to foreign trigram types, exactly as in Subramanya et al. (2010).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Using RandLM and the documented settings (8-bit values and 1 256 false-positive probability), we built a stupid backoff model on the same data as in Section 5.2.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The state function is integrated into the query process so that, in lieu of the query p(wnjwn\u22121 1 ), the application issues query p(wnjs(wn\u22121 1 )) which also returns s(wn1 ).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"7 Acknowledgements.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"3 Throughout this paper we shall give Chinese examples in traditional orthography, followed.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Given a sufficient number of randomly drawn unlabeled examples (i.e., edges), we will induce two completely connected components that together span the entire graph.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"There are two differences between this method and the DL-CoTrain algorithm: spelling and contextual features, alternating between labeling and learning with the two types of features.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"A Hebrew surface token may have several readings, each of which corresponding to a sequence of segments and their corresponding PoS tags.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Lexicon Feature: The string of the token is used as a feature.","label":0},{"source_text":"There is no global pruning.","target_text":"The alignment mapping is j ! i = aj from source position j to target position i = aj . The use of this alignment model raises major problems if a source word has to be aligned to several target words, e.g. when translating German compound nouns.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"As with lexical expections, the semantic classes of co-referring expressions are 4 They may not be perfectly substitutable, for example one NP may be more specific (e.g., \u00e2\u0080\u009che\u00e2\u0080\u009d vs. \u00e2\u0080\u009cJohn F. Kennedy\u00e2\u0080\u009d).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"4 To be sure, it is not always true that a hanzi represents a syllable or that it represents a morpheme.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In (b) is a plausible segmentation for this sentence; in (c) is an implausible segmentation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The ATB gives several different analyses to these words to indicate different types of coordination.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The results of this experiment can be seen in Table 5.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"This problem arises because our keywords consist of only one word.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"63 81.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"input token, the segmentation is then performed deterministically given the 1-best analysis.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"These results are promising and there are several avenues for improving on these results.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"For each extension a new position is added to the coverage set.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, a person\u00e2\u0080\u0099s full name will match with just their last name (e.g., \u00e2\u0080\u009cGeorge Bush\u00e2\u0080\u009d and \u00e2\u0080\u009cBush\u00e2\u0080\u009d), and a company name will match with and without a corporate suffix (e.g., \u00e2\u0080\u009cIBM Corp.\u00e2\u0080\u009d and \u00e2\u0080\u009cIBM\u00e2\u0080\u009d).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"We will refer to our system as MENERGI (Maximum Entropy Named Entity Recognizer using Global Information).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"As noted, this sentence consists of four words, namely B X ri4wen2 'Japanese,' :\u00c2\u00a5, zhanglyu2 'octopus\/ :&P:l zen3me0 'how,' and IDt shuol 'say.'","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"!!:\\ :yu2 e:_nc [::!!:zen3 l!f :moO t:_adv il!:shuot ,:_vb i i i 1 \u00e2\u0080\u00a2 10.03 13...","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Furthermore, even the size of the dictionary per se is less important than the appropriateness of the lexicon to a particular test corpus: as Fung and Wu (1994) have shown, one can obtain substantially better segmentation by tailoring the lexicon to the corpus to be segmented.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Clearly, retaining the original frequencies is important for good performance, and globally smoothing the final weighted frequencies is crucial.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Bikel et al.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This is akin to PoS tags sequences induced by different parses in the setup familiar from English and explored in e.g.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Finally, we model the probability of a new transliterated name as the product of PTN and PTN(hanzi;) for each hanzi; in the putative name.13 The foreign name model is implemented as an WFST, which is then summed with the WFST implementing the dictionary, morpho 13 The current model is too simplistic in several respects.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Unique Occurrences and Zone (UNIQ): This group of features indicates whether the word is unique in the whole document.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Models that employ this strategy are denoted hsp.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We refer to a segment and its assigned PoS tag as a lexeme, and so analyses are in fact sequences of lexemes.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This is a standard adaptation problem for SMT.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In the second scheme, Head+Path, we in addition modify the label of every arc along the lifting path from the syntactic to the linear head so that if the original label is p the new label is p\u2193.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Because many foreign word types are not aligned to an English word (see Table 3), and we do not run label propagation on the foreign side, we expect the projected information to have less coverage.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The authors acknowledge the support of the NSF (CAREER grant IIS0448168, and grant IIS 0904684).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But Rehbein and van Genabith (2007) showed that Evalb should not be used as an indication of real difference\u00e2\u0080\u0094 or similarity\u00e2\u0080\u0094between treebanks.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"These universal POS categories not only facilitate the transfer of POS information from one language to another, but also relieve us from using controversial evaluation metrics,2 by establishing a direct correspondence between the induced hidden states in the foreign language and the observed English labels.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In the numerator, however, the counts of ni1s are quite irregular, in\u00c2\u00ad cluding several zeros (e.g., RAT, none of whose members were seen).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Among these 32 sets, we found the following pairs of sets which have two or more links.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We concentrate on those sets.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This design does not guarantee \u00e2\u0080\u009cstructural zeros,\u00e2\u0080\u009d but biases towards sparsity.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"In general, several gross corpus statistics favor the ATB, so other factors must contribute to parsing underperformance.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The resulting algorithm is depicted in Table 1.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"A corollary of the result of Section 4.3 is that polynomial time recognition of MCTAG's is possible.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For example, if a token starts with a capital letter and ends with a period (such as Mr.), then the feature InitCapPeriod is set to 1, etc. First Word: This feature group contains only one feature firstword.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"First, we identify sources of syntactic ambiguity understudied in the existing parsing literature.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE (without the help of hand-coded systems) has been shown to be somewhat inferior in performance.","label":0},{"source_text":"Here we present two algorithms.","target_text":"This left 962 examples, of which 85 were noise.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"A Person-Prefix-List is compiled in an analogous way.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The horizontal axis in this plot represents the most significant dimension, which explains 62% of the variation.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"It is also worth pointing out a connection with Daum\u00b4e\u2019s (2007) work that splits each feature into domain-specific and general copies.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This is not unreasonable given the application to phrase pairs from OUT, but it suggests that an interesting alternative might be to use a plain log-linear weighting function exp(Ei Aifi(s, t)), with outputs in [0, oo].","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In the next section, we show how an ATM can accept the strings generated by a grammar in a LCFRS formalism in logspace, and hence show that each family can be recognized in polynomial time.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The human evaluators were a non-native, fluent Arabic speaker (the first author) for the ATB and a native English speaker for the WSJ.7 Table 5 shows type- and token-level error rates for each corpus.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This has the potential drawback of increasing the number of features, which can make MERT less stable (Foster and Kuhn, 2009).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"But Rehbein and van Genabith (2007) showed that Evalb should not be used as an indication of real difference\u00e2\u0080\u0094 or similarity\u00e2\u0080\u0094between treebanks.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The verbal reading arises when the maSdar has an NP argument which, in vocalized text, is marked in the accusative case.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"of Tokens MENERGI 318 160,000 200 180,000 IdentiFinder \u00e2\u0080\u0093 650,000 \u00e2\u0080\u0093 790,000 MENE \u00e2\u0080\u0093 \u00e2\u0080\u0093 350 321,000 Table 4: Training Data MUC7 test accuracy.2 For MUC6, the reduction in error due to global features is 27%, and for MUC7,14%.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We adopted the MUC6 guidelines for evaluating coreference relationships based on transitivity in anaphoric chains.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Each lattice arc corresponds to a segment and its corresponding PoS tag, and a path through the lattice corresponds to a specific morphological segmentation of the utterance.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We focus here instead on adapting the two most important features: the language model (LM), which estimates the probability p(wIh) of a target word w following an ngram h; and the translation models (TM) p(slt) and p(t1s), which give the probability of source phrase s translating to target phrase t, and vice versa.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"It uses a maximum entropy framework and classifies each word given its features.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"This might be because our features are more comprehensive than those used by Borthwick.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Thus, provided at least this amount of IN data is available\u2014as it is in our setting\u2014adapting these weights is straightforward.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"In contrast to the Bayesian HMM, \u00ce\u00b8t is not drawn from a distribution which has support for each of the n word types.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The graph was constructed using 2 million trigrams; we chose these by truncating the parallel datasets up to the number of sentence pairs that contained 2 million trigrams.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For instance, the common \"suffixes,\" -nia (e.g.,.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For natural disasters, BABAR generated 20,479 resolutions: 11,652 from lexical seeding and 8,827 from syntactic seeding.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Certainly these linguistic factors increase the difficulty of syntactic disambiguation.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Sequence of Initial Caps (SOIC): In the sentence Even News Broadcasting Corp., noted for its accurate reporting, made the erroneous announcement., a NER may mistake Even News Broadcasting Corp. as an organization name.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The alignment model uses two kinds of parameters: alignment probabilities p(aj jaj\u00f4\u0080\u0080\u00801; I; J), where the probability of alignment aj for position j depends on the previous alignment position aj\u00f4\u0080\u0080\u00801 (Ney et al., 2000) and lexicon probabilities p(fj jeaj ).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The second algorithm extends ideas from boosting algorithms, designed for supervised learning tasks, to the framework suggested by (Blum and Mitchell 98).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For example, if Barry is not in commonWords and is found in the list of person first names, then the feature PersonFirstName will be set to 1.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This technique was introduced by Clarkson and Rosenfeld (1997) and is also implemented by IRSTLM and BerkeleyLM\u2019s compressed option.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"First, we learn weights on individual phrase pairs rather than sentences.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"While the paper mentioned a sorted variant, code was never released.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Later, BerkeleyLM (Pauls and Klein, 2011) described ideas similar to ours.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"Here we present two algorithms.","target_text":", for A. T.&T. nonalpha.. .","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Our code has been publicly available and intergrated into Moses since October 2010.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If these sets do not overlap, then the words cannot be coreferent.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Finally, we show that in application settings, the absence of gold segmentation lowers parsing performance by 2\u00e2\u0080\u00935% F1.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Our primary goal is to exploit the resources that are most appropriate for the task at hand, and our secondary goal is to allow for comparison of our models\u2019 performance against previously reported results.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Finally, we find links between sets of phrases, based on the NE instance pair data (for example, different phrases which link \u00e2\u0080\u009cIBM\u00e2\u0080\u009d and \u00e2\u0080\u009cLotus\u00e2\u0080\u009d) (Step 4).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, in the sentence that starts with \u00e2\u0080\u009cBush put a freeze on . . .","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Two of the Mainlanders also cluster close together but, interestingly, not particularly close to the Taiwan speakers; the third Mainlander is much more similar to the Taiwan speakers.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This information is readily available in TRIE where adjacent records with equal pointers indicate no further extension of context is possible.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(a) ;IE shi4 'be' => ;IE;IE shi4bu2-shi4 (be-not-be) 'is it?'","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Depending on the threshold t0, the search algorithm may miss the globally optimal path which typically results in additional translation errors.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"3.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The work of Thatcher (1973) and Rounds (1969) define formal systems that generate tree sets that are related to CFG's and IG's.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Here, we process only full-form words within the translation procedure.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Full Chinese personal names are in one respect simple: they are always of the form family+given.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In the first part of the experiment, dependency graphs from the treebanks were projectivized using the algorithm described in section 2.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Morphologically derived words such as, xue2shengl+men0.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In particular, it may not be possible to learn functions fi (x f2(x2,t) for i = m + 1...n: either because there is some noise in the data, or because it is just not realistic to expect to learn perfect classifiers given the features used for representation.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"10 Other orthographic normalization schemes have been suggested for Arabic (Habash and Sadat, 2006), but we observe negligible parsing performance differences between these and the simple scheme used in this evaluation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Most languages that use Roman, Greek, Cyrillic, Armenian, or Semitic scripts, and many that use Indian-derived scripts, mark orthographic word boundaries; however, languages written in a Chinese-derived writ\u00c2\u00ad ing system, including Chinese and Japanese, as well as Indian-derived writing systems of languages like Thai, do not delimit orthographic words.1 Put another way, written Chinese simply lacks orthographic words.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The parallel data came from the Europarl corpus (Koehn, 2005) and the ODS United Nations dataset (UN, 2006).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Its success depends on the two domains being relatively close, and on the OUT corpus not being so large as to overwhelm the contribution of IN.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Previous reports on Chinese segmentation have invariably cited performance either in terms of a single percent-correct score, or else a single precision-recall pair.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Maximizing (7) is thus much faster than a typical MERT run. where co(s, t) are the counts from OUT, as in (6).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Consider the case where IX].","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This smooth guarantees that there are no zeroes estimated.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We thank United Informatics for providing us with our corpus of Chinese text, and BDC for the 'Behavior ChineseEnglish Electronic Dictionary.'","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For example, given a sequence F1G1G2, where F1 is a legal single-hanzi family name, and Plural Nouns X g 0 g \"' X X 0 T!i c\"'.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Ltd., then organization will be more probable.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This flexibility, along with the simplicity of implementation and expansion, makes this framework an attractive base for continued research.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In the case of adverbial reduplication illustrated in (3b) an adjective of the form AB is reduplicated as AABB.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For each source word f, the list of its possible translations e is sorted according to p(fje) puni(e), where puni(e) is the unigram probability of the English word e. It is su\u00c3\u0086cient to consider only the best 50 words.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(a) 1 \u00c2\u00a7 . ;m t 7 leO z h e 4 pil m a 3 lu 4 sh an g4 bi ng 4 t h i s CL (assi fier) horse w ay on sic k A SP (ec t) 'This horse got sick on the way' (b) 1\u00c2\u00a7: . til y zhe4 tiao2 ma3lu4 hen3 shao3 this CL road very few 'Very few cars pass by this road' :$ chel jinglguo4 car pass by 2.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"(b) After they were released...","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The method just described segments dictionary words, but as noted in Section 1, there are several classes of words that should be handled that are not found in a standard dictionary.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Le m\u00b4edicament de r\u00b4ef\u00b4erence de Silapo est EPREX\/ERYPO, qui contient de l\u2019\u00b4epo\u00b4etine alfa.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Our TRIE implements the popular reverse trie, in which the last word of an n-gram is looked up first, as do SRILM, IRSTLM\u2019s inverted variant, and BerkeleyLM except for the scrolling variant.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For alif with hamza, normalization can be seen as another level of devocalization.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"However, this argument is only plausible if the formal framework allows non-projective dependency structures, i.e. structures where a head and its dependents may correspond to a discontinuous constituent.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"This kind of dependency arises from the use of the composition operation to compose two arbitrarily large categories.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"30 75.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The average agreement among the human judges is .76, and the average agreement between ST and the humans is .75, or about 99% of the interhuman agreement.15 One can better visualize the precision-recall similarity matrix by producing from that matrix a distance matrix, computing a classical metric multidimensional scaling (Torgerson 1958; Becker, Chambers, Wilks 1988) on that dis\u00c2\u00ad tance matrix, and plotting the first two most significant dimensions.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"to explore how well we can induce POS tags using only the one-tag-per-word constraint.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"There are thus some very good reasons why segmentation into words is an important task.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In this paper, we show that the maximum entropy framework is able to make use of global information directly, and achieves performance that is comparable to the best previous machine learning-based NERs on MUC6 and MUC7 test data.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"However, the data sparsity induced by vocalization makes it difficult to train statistical models on corpora of the size of the ATB, so vocalizing and then parsing may well not help performance.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"However, it is almost universally the case that no clear definition of what constitutes a \"correct\" segmentation is given, so these performance measures are hard to evaluate.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In this case, we have no finite-state restrictions for the search space.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"However, the data sparsity induced by vocalization makes it difficult to train statistical models on corpora of the size of the ATB, so vocalizing and then parsing may well not help performance.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We tabulate this increase in Table 3.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The final estimating equation is then: (3) Since the total of all these class estimates was about 10% off from the Turing estimate n1\/N for the probability of all unseen hanzi, we renormalized the estimates so that they would sum to n 1jN.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Evaluation metrics used are Attachment Score (AS), i.e. the proportion of tokens that are attached to the correct head, and Exact Match (EM), i.e. the proportion of sentences for which the dependency graph exactly matches the gold standard.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The model is often further restricted so that each source word is assigned to exactly one target word (Brown et al., 1993; Ney et al., 2000).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"For example, we may have semantic information (e.g. database query operations) associated with the productions in a grammar.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We used the TF\/ITF metric to identify keywords.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The availability of these resources guided our selection of foreign languages.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"For each feature type f and tag t, a multinomial \u00cf\u0088tf is drawn from a symmetric Dirichlet distribution with concentration parameter \u00ce\u00b2.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"More formally, we start by representing the dictionary D as a Weighted Finite State Trans\u00c2\u00ad ducer (WFST) (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","label":1},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"For parsing, the most challenging form of ambiguity occurs at the discourse level.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Figure 4 shows a constituent headed by a process nominal with an embedded adjective phrase.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"It has been shown for English (Wang and Hirschberg 1992; Hirschberg 1993; Sproat 1994, inter alia) that grammatical part of speech provides useful information for these tasks.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For all variants, we found that BerkeleyLM always rounds the floating-point mantissa to 12 bits then stores indices to unique rounded floats.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This implies, therefore, that a major factor in the performance of a Chinese segmenter is the quality of the base dictionary, and this is probably a more important factor-from the point of view of performance alone-than the particular computational methods used.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Consequently, we implemented our own annotation tool ConAno in Java (Stede, Heintze 2004), which provides specifically the functionality needed for our purpose.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"BABAR employs information extraction techniques to represent and learn role relationships.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We use two different similarity functions to define the edge weights among the foreign vertices and between vertices from different languages.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"It is striking that from this point of view many formalisms can be grouped together as having identically structured derivation tree sets.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"That is, we can use the discourse parser on PCC texts, emulating for instance a \u00e2\u0080\u009cco-reference oracle\u00e2\u0080\u009d that adds the information from our co-reference annotations.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The number of top-ranked pairs to retain is chosen to optimize dev-set BLEU score.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The probabilistic version of this procedure is straightforward: We once again assume independence among our various member parsers.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"It reads a file with a list of German connectives, and when a text is opened for annotation, it highlights all the words that show up in this list; these will be all the potential connectives.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"In order to evaluate and advance this approach, it helps to feed into the knowledge base data that is already enriched with some of the desired information \u00e2\u0080\u0094 as in PCC.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Pairwise comparison is done using the sign test.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Step 2.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We show the results of three of the experiments we conducted to measure isolated constituent precision under various partitioning schemes.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"They are also labelled for their topicality (yes \/ no), and this annotation is accompanied by a confidence value assigned by the annotator (since it is a more subjective matter).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We address the question of whether or not a formalism can generate only structural descriptions with independent paths.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"First of all, most previous articles report perfor\u00c2\u00ad mance in terms of a single percent-correct score, or else in terms of the paired measures of precision and recall.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We use Alternating Turing Machines (Chandra, Kozen, and Stockmeyer, 1981) to show that polynomial time recognition is possible for the languages discussed in Section 4.3.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We generate these caseframes automatically by running AutoSlog over the training corpus exhaustively so that it literally generates a pattern to extract every noun phrase in the corpus.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"At each iteration the algorithm increases the number of rules, while maintaining a high level of agreement between the spelling and contextual decision lists.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Precision.","label":0},{"source_text":"There is no global pruning.","target_text":"In general, m; l; l0 6= fl1; l2; l3g and in line umber 3 and 4, l0 must be chosen not to violate the above reordering restriction.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(If fewer than n rules have Precision greater than pin, we 3Note that taking tlie top n most frequent rules already makes the method robut to low count events, hence we do not use smoothing, allowing low-count high-precision features to be chosen on later iterations. keep only those rules which exceed the precision threshold.) pm,n was fixed at 0.95 in all experiments in this paper.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We also performed experiments to evaluate the impact of each type of contextual role knowledge separately.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"In both cases, the instanceweighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline, and gains of between 0.6 and 1.8 over an equivalent mixture model (with an identical training procedure but without instance weighting).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"An error count of 0:0 is assigned to a perfect translation, and an error count of 1:0 is assigned to a semantically and syntactically wrong translation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"97 78.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"There are 13 types of extensions needed to describe the verbgroup reordering.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"A geometrical progression of language families defined by Weir (1987) involves tree sets with increasingly complex path sets.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Because many foreign word types are not aligned to an English word (see Table 3), and we do not run label propagation on the foreign side, we expect the projected information to have less coverage.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"For each language and setting, we report one-to-one (11) and many- to-one (m-1) accuracies.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Two subjects are each given a calendar and they are asked to schedule a meeting.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Many morphological decisions are based on long distance dependencies, and when the global syntactic evidence disagrees with evidence based on local linear context, the two models compete with one another, despite the fact that the PCFG takes also local context into account.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"95 76.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Minimal perfect hashing is used to find the index at which a quantized probability and possibly backoff are stored.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In most cases, however, these expansions come with a steep increase in model complexity, with respect to training procedure and inference time.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Across all languages, +PRIOR consistently outperforms 1TW, reducing error on average by 9.1% and 5.9% on best and median settings respectively.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Thus the method makes the fairly strong assumption that the features can be partitioned into two types such that each type alone is sufficient for classification.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Chunking is not enough to find such relationships.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"With respect to exact match, the improvement is even more noticeable, which shows quite clearly that even if non-projective dependencies are rare on the token level, they are nevertheless important for getting the global syntactic structure correct.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Its correct antecedent is \u00e2\u0080\u009ca revolver\u00e2\u0080\u009d, which is extracted by the caseframe \u00e2\u0080\u009ckilled with <NP>\u00e2\u0080\u009d.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"T e r r o r i s m Ca sef ra me Semantic Classes <a ge nt > ass ass ina ted group, human inv esti gat ion int o < N P> event exp lod ed out sid e < N P> building N a t u r a l D i s a s t e r s Ca sef ra me Semantic Classes <a ge nt > inv esti gat ing cau se group, human sur viv or of < N P> event, natphenom hit wit h < N P> attribute, natphenom Figure 3: Semantic Caseframe Expectations Figure 2: Lexical Caseframe Expectations To illustrate how lexical expectations are used, suppose we want to determine whether noun phrase X is the antecedent for noun phrase Y. If they are coreferent, then X and Y should be substitutable for one another in the story.4 Consider these sentences: (S1) Fred was killed by a masked man with a revolver.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We must adjoin all trees in an auxiliary tree set together as a single step in the derivation.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The quasi-monotone search performs best in terms of both error rates mWER and SSER.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Using RandLM and the documented settings (8-bit values and 1 256 false-positive probability), we built a stupid backoff model on the same data as in Section 5.2.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"At each point during the derivation, the prediction is based on six word tokens, the two topmost tokens on the stack, and the next four input tokens.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"A Person-Prefix-List is compiled in an analogous way.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We considered using the MUC6 and MUC7 data sets, but their training sets were far too small to learn reliable co-occurrence statistics for a large set of contextual role relationships.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Special thanks to Jan Haji\u02c7c and Matthias Trautner Kromann for assistance with the Czech and Danish data, respectively, and to Jan Haji\u02c7c, Tom\u00b4a\u02c7s Holan, Dan Zeman and three anonymous reviewers for valuable comments on a preliminary version of the paper.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Kollege.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In section 2 we introduce the graph transformation techniques used to projectivize and deprojectivize dependency graphs, and in section 3 we describe the data-driven dependency parser that is the core of our system.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"This can be repeated several times to collect a list of author \/ book title pairs and expressions.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"!!:\\ :yu2 e:_nc [::!!:zen3 l!f :moO t:_adv il!:shuot ,:_vb i i i 1 \u00e2\u0080\u00a2 10.03 13...","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"In Equations 1 through 3 we develop the model for constructing our parse using na\u00efve Bayes classification.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Following the setup of Johnson (2007), we use the whole of the Penn Treebank corpus for training and evaluation on English.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Arabic is a morphologically rich language with a root-and-pattern system similar to other Semitic languages.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The learned information was recycled back into the resolver to improve its performance.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The Levenshtein distance between the automatic translation and each of the reference translations is computed, and the minimum Levenshtein distance is taken.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Specifically, we assume each word type W consists of feature-value pairs (f, v).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, if a token starts with a capital letter and ends with a period (such as Mr.), then the feature InitCapPeriod is set to 1, etc. First Word: This feature group contains only one feature firstword.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"After the first step towards breadth had been taken with the PoS-tagging, RST annotation, and URML conversion of the entire corpus of 170 texts12 , emphasis shifted towards depth.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The learning task is to find two classifiers : 2x1 { \u20141, +1} 12 : 2x2 { \u20141, +1} such that (x1,) = f2(x2,t) = yt for examples i = 1, , m, and f1 (x1,) = f2 (x2,t) as often as possible on examples i = m + 1, ,n. To achieve this goal we extend the auxiliary function that bounds the training error (see Equ.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Cohen and Smith (2007) later on based a system for joint inference on factored, independent, morphological and syntactic components of which scores are combined to cater for the joint inference task.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The development of automatic scoring methods is an open field of research.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This is most severe with RandLM in the multi-threaded case, where each thread keeps a separate cache, exceeding the original model size.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Ex: The brigade, which attacked ...","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"But we also need an estimate of the probability for a non-occurring though possible plural form like i\u00c2\u00a5JJ1l.f, nan2gua1-men0 'pumpkins.'","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"(2006) developed a technique for splitting and chunking long sentences.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"When OUT is large and distinct, its contribution can be controlled by training separate IN and OUT models, and weighting their combination.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"It can also be seen clearly in this plot that two of the Taiwan speakers cluster very closely together, and the third Tai\u00c2\u00ad wan speaker is also close in the most significant dimension (the x axis).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Then we ran binary search to determine the least amount of memory with which it would run.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Therefore in cases where the segmentation is identical between the two systems we assume that tagging is also identical.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The final block in table 2 shows models trained on feature subsets and on the SVM feature described in 3.4.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The first issue relates to the completeness of the base lexicon.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Consider the following sentences: (a) Jose Maria Martinez, Roberto Lisandy, and Dino Rossy, who were staying at a Tecun Uman hotel, were kidnapped by armed men who took them to an unknown place.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We can obtain a letter equivalent CFL defined by a CFG in which the for each rule as above, we have the production A \u2014* A1 Anup where tk (up) = cp.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"newspaper material, but also including kungfu fiction, Buddhist tracts, and scientific material.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"While RST (Mann, Thompson 1988) proposed that a single relation hold between adjacent text segments, SDRT (Asher, Lascarides 2003) maintains that multiple relations may hold simultaneously.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"besuchen 9.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We thank members of the MIT NLP group for their suggestions and comments.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Most languages that use Roman, Greek, Cyrillic, Armenian, or Semitic scripts, and many that use Indian-derived scripts, mark orthographic word boundaries; however, languages written in a Chinese-derived writ\u00c2\u00ad ing system, including Chinese and Japanese, as well as Indian-derived writing systems of languages like Thai, do not delimit orthographic words.1 Put another way, written Chinese simply lacks orthographic words.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We consider two variants of Berg-Kirkpatrick et al.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We use a simple TF\/IDF method to measure the topicality of words.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Statistics for all data sets are shown in Table 2.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Realizing gains in practice can be challenging, however, particularly when the target domain is distant from the background data.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"conceptual relationship in the discourse.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In this case, we have no finite-state restrictions for the search space.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Instead, we want to apply an inverse transformation to recover the underlying (nonprojective) dependency graph.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Parameter Component As in the standard Bayesian HMM (Goldwater and Griffiths, 2007), all distributions are independently drawn from symmetric Dirichlet distributions: 2 Note that t and w denote tag and word sequences respectively, rather than individual tokens or tags.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"6. 3) all G o l d P O S 7 0 0.7 91 0.825 358 0.7 73 0.818 358 0.8 02 0.836 452 80.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We paid particular attention to minimize the number of free parameters, and used the same hyperparameters for all language pairs, rather than attempting language-specific tuning.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Performance typically stabilizes across languages after only a few number of iterations.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The algorithm can be viewed as heuristically optimizing an objective function suggested by (Blum and Mitchell 98); empirically it is shown to be quite successful in optimizing this criterion.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Using the concept of inverted alignments, we explicitly take care of the coverage constraint by introducing a coverage set C of source sentence positions that have been already processed.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In section 5, we then evaluate the entire parsing system by training and evaluating on data from the Prague Dependency Treebank.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We asked six native speakers-three from Taiwan (TlT3), and three from the Mainland (M1M3)-to segment the corpus.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The task is to learn a function from an input string (proper name) to its type, which we will assume to be one of the categories Person, Organization, or Location.","label":0},{"source_text":"Here we present two algorithms.","target_text":"2.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Again, famous place names will most likely be found in the dictionary, but less well-known names, such as 1PM\u00c2\u00b1 R; bu4lang3-shi4wei2-ke4 'Brunswick' (as in the New Jersey town name 'New Brunswick') will not generally be found.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Zone: MUC data contains SGML tags, and a document is divided into zones (e.g., headlines and text zones).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In this section, we brie y review our translation approach.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"We have used the Java-based opennlp maximum entropy package1.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"5.1 Data Sets.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We can model this probability straightforwardly enough with a probabilistic version of the grammar just given, which would assign probabilities to the individual rules.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"In our experiments, we used the same set of features as BergKirkpatrick et al. (2010): an indicator feature based In a traditional Markov model, the emission distribution P\u0398(Xi = xi  |Zi = zi) is a set of multinomials.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For definite NPs, the results are a mixed bag: some knowledge sources increased recall a little, but at the expense of some precision.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For example, in Northern dialects (such as Beijing), a full tone (1, 2, 3, or 4) is changed to a neutral tone (0) in the final syllable of many words: Jll donglgual 'winter melon' is often pronounced donglguaO.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"So we decided to use semantic class information only to rule out candidates.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In an experiment on automatic rhetorical parsing, the RST-annotations and PoS tags were used by (Reitter 2003) as a training corpus for statistical classification with Support Vector Machines.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For the experiments, we use a simple preprocessing step.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Therefore, we want state to encode the minimum amount of information necessary to properly compute language model scores, so that the decoder will be faster and make fewer search errors.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The standard measures for evaluating Penn Treebank parsing performance are precision and recall of the predicted constituents.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"These performance gains transfer to improved system runtime performance; though we focused on Moses, our code is the best lossless option with cdec and Joshua.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"68 96.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"1 55.8 38.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"When two partial hypotheses have equal state (including that of other features), they can be recombined and thereafter efficiently handled as a single packed hypothesis.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"2.2.1 The Caseframe Representation Information extraction (IE) systems use extraction patterns to identify noun phrases that play a specific role in 1 Our implementation only resolves NPs that occur in the same document, but in retrospect, one could probably resolve instances of the same existential NP in different documents too.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We briefly review it here for completeness.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Ltd., then organization will be more probable.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"What is important and is not expressed by the notation is the so-called coverage constraint: each source position j should be 'hit' exactly once by the path of the inverted alignment bI 1 = b1:::bi:::bI . Using the inverted alignments in the maximum approximation, we obtain as search criterion: max I (p(JjI) max eI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) max bI 1 I Yi=1 [p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])) = = max I (p(JjI) max eI 1;bI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])); where the two products over i have been merged into a single product over i. p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) is the trigram language model probability.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"All of these systems were run on data that was not seen during their development.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"43 80.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"For parsing, this is a mistake, especially in the case of interrogatives.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Morphological Analyzer Ideally, we would use an of-the-shelf morphological analyzer for mapping each input token to its possible analyses.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This suggests a strategy: run interpolation search until the range narrows to 4096 or fewer entries, then switch to binary search.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"P St = n. \u00ce\u00b2 T VARIABLES \u00cf\u0088 Y W : Word types (W1 ,.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Named Entity Recognition: A Maximum Entropy Approach Using Global Information","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For each caseframe, BABAR collects the semantic classes associated with the head nouns of NPs that were extracted by the caseframe.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Also, the argument has been made that machine translation performance should be evaluated via task-based evaluation metrics, i.e. how much it assists performing a useful task, such as supporting human translators or aiding the analysis of texts.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In IE, creating the patterns which express the requested scenario, e.g. \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d or \u00e2\u0080\u009ccorporate merger and acquisition\u00e2\u0080\u009d is regarded as the hardest task.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We found that contextual role knowledge was more beneficial for pronouns than for definite noun phrases.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Several extensions of AdaBoost for multiclass problems have been suggested (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Note that in line 4 the last visited position for the successor hypothesis must be m. Otherwise , there will be four uncovered positions for the predecessor hypothesis violating the restriction.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The problem is a binary classification problem.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"This alters generation of T as follows: n P (T |\u00cf\u0088) = n P (Ti|\u00cf\u0088) i=1 Note that this distribution captures the frequency of a tag across word types, as opposed to tokens.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"the number of permutations carried out for the word reordering is given.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In the CC-domain, there are 32 sets of phrases which contain more than 2 phrases.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Vocabulary lookup is a hash table mapping from word to vocabulary index.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Our \u201cProjection\u201d baseline is able to benefit from the bilingual information and greatly improves upon the monolingual baselines, but falls short of the \u201cNo LP\u201d model by 2.5% on an average.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Lemma: If the number of votes required by constituent voting is greater than half of the parsers under consideration the resulting structure has no crossing constituents.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The tool we use is MMAX8, which has been specifically designed for marking co-reference.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The structure uses linear probing hash tables and is designed for speed.","label":0},{"source_text":"Here we present two algorithms.","target_text":"AdaBoost.MH can be applied to the problem using these pseudolabels in place of supervised examples.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The theory has also been validated empirically.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"In this paper we study the problem of using a parallel corpus from a background domain (OUT) to improve performance on a target domain (IN) for which a smaller amount of parallel training material\u2014though adequate for reasonable performance\u2014is also available.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"6 Our knowledge sources return some sort of probability estimate, although in some cases this estimate is not especially well-principled (e.g., the Recency KS).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For the seen word ir, 'gen\u00c2\u00ad erals,' there is an c:NC transduction from to the node preceding ir,; this arc has cost cost( f,) - cost(unseen(f,)), so that the cost of the whole path is the desired cost( f,).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"N, portion of examples on which both classifiers give a label rather than abstaining), and the proportion of these examples on which the two classifiers agree.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The latter arcs correspond to OOV words in English.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Inspection of the data shows that at n = 2500, the two classifiers both give labels on 44,281 (49.2%) of the unlabeled examples, and give the same label on 99.25% of these cases.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We used 22 features for the logistic weighting model, divided into two groups: one intended to reflect the degree to which a phrase pair belongs to general language, and one intended to capture similarity to the IN domain.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"To initialize the graph we tag the English side of the parallel text using a supervised model.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Table 2 shows single-threaded results, mostly for comparison to IRSTLM, and Table 3 shows multi-threaded results.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Consider first the examples in (2).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Subsets of partial hypotheses with coverage sets C of increasing cardinality c are processed.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"For 2 < n < N, we use a hash table mapping from the n-gram to the probability and backoff3.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"constitute names, since we have only their segmentation, not the actual classification of the segmented words.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"We then evaluate the approach in two steps.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Machine Translation In this case my colleague can not visit on I n d i e s e m F a l l ka nn m e i n K o l l e g e a m the v i e r t e n M a i n i c h t b e s u c h e n S i e you fourth of May Figure 1: Reordering for the German verbgroup.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In this paper we have argued that Chinese word segmentation can be modeled ef\u00c2\u00ad fectively using weighted finite-state transducers.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Furthermore, we know one of the original parses will be the hypothesized parse, so the direct method of determining which one is best is to compute the probability of each of the candidate parses using the probabilistic model we developed in Section 2.1.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The edge from the root to the subtree for the derivation of 7i is labeled by the address ni.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"On average, 6 reference translations per automatic translation are available.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"On several languages, we report performance exceeding that of state-of-the art systems.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"By definition, each existential NP uniquely specifies an object or concept, so we can infer that all instances of the same existential NP are coreferent (e.g., \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d always refers to the same entity).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"0 Figure 5 An example of affixation: the plural affix.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"3.1 Gross Statistics.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"5 \u00e2\u0080\u0098Underspecified Rhetorical Markup Language\u00e2\u0080\u0099 6 This confirms the figure given by (Schauer, Hahn.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"An alternate approximation to (8) would be to let w,\\(s, t) directly approximate p\u02c6I(s, t).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The cost is computed as follows, where N is the corpus size and f is the frequency: (1) Besides actual words from the base dictionary, the lexicon contains all hanzi in the Big 5 Chinese code\/ with their pronunciation(s), plus entries for other characters that can be found in Chinese text, such as Roman letters, numerals, and special symbols.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Given that part-of-speech labels are properties of words rather than morphemes, it follows that one cannot do part-of-speech assignment without having access to word-boundary information.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The traveling salesman problem is an optimization problem which is defined as follows: given are a set of cities S = s1; ; sn and for each pair of cities si; sj the cost dij > 0 for traveling from city si to city sj . We are looking for the shortest tour visiting all cities exactly once while starting and ending in city s1.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"of Articles No.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"It may be more realistic to replace the second criteria with a softer one, for example (Blum and Mitchell 98) suggest the alternative Alternatively, if Ii and 12 are probabilistic learners, it might make sense to encode the second constraint as one of minimizing some measure of the distance between the distributions given by the two learners.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"And indeed, converging on annotation guidelines is even more difficult than it is with co-reference.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In turn we use two sorts of heuristics, orthogonal to one another, to prune segmentation possibilities based on lexical and grammatical constraints.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The English data comes from the WSJ portion of the Penn Treebank and the other languages from the training set of the CoNLL-X multilingual dependency parsing shared task.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Two strings are letter equivalent if they contain equal number of occurrences of each terminal symbol, and two languages are letter equivalent if every string in one language is letter equivalent to a string in the other language and vice-versa.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The probability distribution that satisfies the above property is the one with the highest entropy.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, the passive voice pattern \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d and the active voice pattern \u00e2\u0080\u009ckidnapped <direct object>\u00e2\u0080\u009d are merged into a single normalized pattern \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d.2 For the sake of sim plicity, we will refer to these normalized extraction patterns as caseframes.3 These caseframes can capture two types of contextual role information: (1) thematic roles corresponding to events (e.g, \u00e2\u0080\u009c<agent> kidnapped\u00e2\u0080\u009d or \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d), and (2) predicate-argument relations associated with both verbs and nouns (e.g., \u00e2\u0080\u009ckidnapped for <np>\u00e2\u0080\u009d or \u00e2\u0080\u009cvehicle with <np>\u00e2\u0080\u009d).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Thus corresponding pseudo-labels for instances on which gj abstain are set to zero and these instances do not contribute to the objective function.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Nonetheless, the results of the comparison with human judges demonstrates that there is mileage being gained by incorporating models of these types of words.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This analyzer setting is similar to that of (Cohen and Smith, 2007), and models using it are denoted nohsp, Parser and Grammar We used BitPar (Schmid, 2004), an efficient general purpose parser,10 together with various treebank grammars to parse the input sentences and propose compatible morphological segmentation and syntactic analysis.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"input: source string f1:::fj :::fJ initialization for each cardinality c = 1; 2; ; J do for each pair (C; j), where j 2 C and jCj = c do for each target word e 2 E Qe0 (e; C; j) = p(fj je) max \u00c3\u0086;e00 j02Cnfjg fp(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j0)g words fj in the input string of length J. For the final translation each source position is considered exactly once.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"So, 1: f, xue2shengl+men0 (student+PL) 'students' occurs and we estimate its cost at 11.43; similarly we estimate the cost of f, jiang4+men0 (general+PL) 'generals' (as in 'J' f, xiao3jiang4+men0 'little generals'), at 15.02.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Decreasing the threshold results in higher mWER due to additional search errors.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"This paper proposes a simple and effective tagging method that directly models tag sparsity and other distributional properties of valid POS tag assignments.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Making the reasonable assumption that similar information is relevant for solving these problems in Chinese, it follows that a prerequisite for intonation-boundary assignment and prominence assignment is word segmentation.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The general-language features have a slight advantage over the similarity features, and both are better than the SVM feature.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"To do this, the x's and y's are stored in the next 2ni + 2n2 tapes, and M goes to a universal state.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Preferences for constituent order (especially in languages with relatively free word order) often belong to this group.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In contrast, our morphological probabilities are based on a unigram, lexeme-based model, and all other (local and non-local) contextual considerations are delegated to the PCFG.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In the named entity task, X1 might be the instance space for the spelling features, X2 might be the instance space for the contextual features.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"a classifier.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Bikel et al.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Zone: MUC data contains SGML tags, and a document is divided into zones (e.g., headlines and text zones).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We have presented a method for unsupervised part- of-speech tagging that considers a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In recent years, coreference resolvers have been evaluated as part of MUC6 and MUC7 (MUC7 Proceedings, 1998).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"It is formally straightforward to extend the grammar to include these names, though it does increase the likelihood of overgeneration and we are unaware of any working systems that incorporate this type of name.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Standard SMT systems have a hierarchical parameter structure: top-level log-linear weights are used to combine a small set of complex features, interpreted as log probabilities, many of which have their own internal parameters and objectives.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In (Bikel et al., 1997) and (Bikel et al., 1999), performance was plotted against training data size to show how performance improves with training data size.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"One conclusion drawn from this annotation effort was that for humans and machines alike, 2 www.sfs.nphil.unituebingen.de\/Elwis\/stts\/ stts.html 3 www.coli.unisb.de\/sfb378\/negra-corpus\/annotate.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The entries in such a lexicon may be thought of as meaningful surface segments paired up with their PoS tags li = (si, pi), but note that a surface segment s need not be a space-delimited token.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Consider for a set of constituents the isolated constituent precision parser metric, the portion of isolated constituents that are correctly hypothesized.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Due to the dramatic fiscal situation in Brandenburg she now surprisingly withdrew legislation drafted more than a year ago, and suggested to decide on it not before 2003.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Otherwise for the predecessor search hypothesis, we would have chosen a position that would not have been among the first n uncovered positions.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The same argument holds for resolving PP attachment of a prefixed preposition or marking conjunction of elements of any kind.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"0 57.2 43.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"MENE is then trained on 80% of the training corpus, and tested on the remaining 20%.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(If fewer than n rules have Precision greater than pin, we 3Note that taking tlie top n most frequent rules already makes the method robut to low count events, hence we do not use smoothing, allowing low-count high-precision features to be chosen on later iterations. keep only those rules which exceed the precision threshold.) pm,n was fixed at 0.95 in all experiments in this paper.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The unlabeled data gives many such &quot;hints&quot; that two features should predict the same label, and these hints turn out to be surprisingly useful when building a classifier.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"For example, one of the ATB samples was the determiner -\"\" ; dhalik\u00e2\u0080\u009cthat.\u00e2\u0080\u009d The sample occurred in 1507 corpus po sitions, and we found that the annotations were consistent.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If is seen infrequently during training (less than a small count), then will not be selected as a feature and all features in this group are set to 0.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Of course, since the number of attested (phonemic) Mandarin syllables (roughly 1400, including tonal distinctions) is far smaller than the number of morphemes, it follows that a given syllable could in principle be written with any of several different hanzi, depending upon which morpheme is intended: the syllable zhongl could be lfl 'middle,''clock,''end,' or ,'loyal.'","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"both MENE and IdentiFinder used more training data than we did (we used only the official MUC 6 and MUC7 training data).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"One of such approaches uses comparable documents, which are sets of documents whose content are found\/known to be almost the same, such as different newspaper stories about the same event [Shinyama and Sekine 03] or different translations of the same story [Barzilay 01].","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We show the results of three of the experiments we conducted to measure isolated constituent precision under various partitioning schemes.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Adding the isolated constituents to our hypothesis parse could increase our expected recall, but in the cases we investigated it would invariably hurt our precision more than we would gain on recall.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"This approach is compared to another reordering scheme presented in (Berger et al., 1996).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"For this reason, na\u00efve Bayes classifiers are well-matched to this problem.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The F- measure score increased for both domains, reflecting a substantial increase in recall with a small decrease in precision.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(Specifically, the limit n starts at 5 and increases by 5 at each iteration.)","label":0},{"source_text":"Here we present two algorithms.","target_text":"Each vertex within a connected component must have the same label \u2014 in the binary classification case, we need a single labeled example to identify which component should get which label.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Hence, their relationship to formalisms such as HG's and TAG's is of interest.","label":0},{"source_text":"Here we present two algorithms.","target_text":"1,000 of these were picked at random, and labeled by hand to produce a test set.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"During the search process, a partial hypothesis is extended by choosing a source sentence position, which has not been aligned with a target sentence position yet.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As noted, this sentence consists of four words, namely B X ri4wen2 'Japanese,' :\u00c2\u00a5, zhanglyu2 'octopus\/ :&P:l zen3me0 'how,' and IDt shuol 'say.'","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":".., Tn ) T W \u00cf\u0086 E w : Token word seqs (obs) t : Token tag assigns (det by T ) PARAMETERS \u00cf\u0088 : Lexicon parameters \u00ce\u00b8 : Token word emission parameters \u00cf\u0086 : Token tag transition parameters \u00cf\u0086 \u00cf\u0086 t1 t2 \u00ce\u00b8 \u00ce\u00b8 w1 w2 K \u00cf\u0086 T tm O K \u00ce\u00b8 E wN m N N Figure 1: Graphical depiction of our model and summary of latent variables and parameters.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This step ensures that the most frequent terms for each domain are labeled (in case some of them are not in WordNet) and labeled with the sense most appropriate for the domain.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For pronouns, however, all of the knowledge sources increased recall, often substantially, and with little if any decrease in precision.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The use of global features has improved the performance on MUC6 test data from 90.75% to 93.27% (27% reduction in errors), and the performance on MUC7 test data from 85.22% to 87.24% (14% reduction in errors).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We use the HSPELL9 (Har\u2019el and Kenigsberg, 2004) wordlist as a lexeme-based lexicon for pruning segmentations involving invalid segments.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We briefly review it here for completeness.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Unsupervised learning approaches appear to be a natural solution to this problem, as they require only unannotated text for training models.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"But in most cases they can be used interchangably.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Of these cases, 38 were temporal expressions (either a day of the week or month of the year).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"3.4 Salience-based text generation.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Their results are then compared with the results of an automatic segmenter.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"However, this optimistic search would not visit the entries necessary to store backoff information in the outgoing state.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"+ cost(unseen(fm, as desired.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Now we have sets of phrases which share a keyword and we have links between those sets.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"As we have noted in Section 2, the general semantic class to which a hanzi belongs is often predictable from its semantic radical.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"18 We are grateful to ChaoHuang Chang for providing us with this set.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Step 4.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The morphological anal\u00c2\u00adysis itself can be handled using well-known techniques from finite-state morphol 9 The initial estimates are derived from the frequencies in the corpus of the strings of hanzi making up.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But it conflates the coordinating and discourse separator functions of wa (<..4.b \u00ef\u00bf\u00bd \u00ef\u00bf\u00bd) into one analysis: conjunction(Table 3).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The computation of Pfr1(c)1Mi M k (C)) has been sketched before in Equations 1 through 4.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"So, 1: f, xue2shengl+men0 (student+PL) 'students' occurs and we estimate its cost at 11.43; similarly we estimate the cost of f, jiang4+men0 (general+PL) 'generals' (as in 'J' f, xiao3jiang4+men0 'little generals'), at 15.02.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"It can be easily shown from Thatcher's result that the path set of every local set is a regular set.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"On a set of 11 sentence fragments-the A set-where they reported 100% recall and precision for name identification, we had 73% recall and 80% precision.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"segmentation (Table 2).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The second experiment is limited to data from PDT.5 The training part of the treebank was projectivized under different encoding schemes and used to train memory-based dependency parsers, which were run on the test part of the treebank, consisting of 7,507 sentences and 125,713 tokens.6 The inverse transformation was applied to the output of the parsers and the result compared to the gold standard test set.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"9 66.4 47.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"In our experiments, we used the same set of features as BergKirkpatrick et al. (2010): an indicator feature based In a traditional Markov model, the emission distribution P\u0398(Xi = xi  |Zi = zi) is a set of multinomials.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Section 3 discusses the applications that have been completed with PCC, or are under way, or are planned for the future.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For the examples given in (1) and (2) this certainly seems possible.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"A Person-Prefix-List is compiled in an analogous way.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The token-level term is similar to the standard HMM sampling equations found in Johnson (2007).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"On the other hand, if it is seen as McCann Pte.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"To investigate the influence of these factors, we analyze Modern Standard Arabic (henceforth MSA, or simply \u00e2\u0080\u009cArabic\u00e2\u0080\u009d) because of the unusual opportunity it presents for comparison to English parsing results.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"First, we learn weights on individual phrase pairs rather than sentences.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.2 Syntactic structure.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Therefore, for n-gram wn1 , all leftward extensions wn0 are an adjacent block in the n + 1-gram array.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Segmentation errors cascade into the parsing phase, placing an artificial limit on parsing performance.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The general-language features have a slight advantage over the similarity features, and both are better than the SVM feature.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We focus on phrases which connect two Named Entities (NEs), and proceed in two stages.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This is motivated by taking \u03b2 po(s|t) to be the parameters of a Dirichlet prior on phrase probabilities, then maximizing posterior estimates p(s|t) given the IN corpus.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In this paper, we show that the maximum entropy framework is able to make use of global information directly, and achieves performance that is comparable to the best previous machine learning-based NERs on MUC6 and MUC7 test data.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The maximum entropy framework estimates probabilities based on the principle of making as few assumptions as possible, other than the constraints imposed.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"In the label propagation stage, we propagate the automatic English tags to the aligned Italian trigram types, followed by further propagation solely among the Italian vertices. the Italian vertices are connected to an automatically labeled English vertex.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For all languages, the vocabulary sizes increase by several thousand words.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Je voudrais pr\u00b4eciser, a` l\u2019adresse du commissaire Liikanen, qu\u2019il n\u2019est pas ais\u00b4e de recourir aux tribunaux nationaux.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"1 The apparent difficulty of adapting constituency models to non-configurational languages has been one motivation for dependency representations (Hajic\u00cb\u0087 and Zema\u00c2\u00b4nek, 2004; Habash and Roth, 2009).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"In general, the neighborhoods can be more diverse and we allow a soft label distribution over the vertices.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"An edge indicates that the two features must have the same label.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Another option is the closedsource data structures from Sheffield (Guthrie and Hepple, 2010).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"One striking example is the error reduction for Spanish, which reduces error by 36.5% and 24.7% for the best and median settings respectively.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"La ng ua ge # To ke ns # W or d Ty pe s # Ta gs E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 1 1 7 3 7 6 6 9 4 3 8 6 2 0 3 5 6 8 6 9 9 6 0 5 2 0 6 6 7 8 8 9 3 3 4 1 9 1 4 6 7 4 9 2 0 6 1 8 3 5 6 2 8 3 9 3 7 2 3 2 5 2 8 9 3 1 1 6 4 5 8 2 0 0 5 7 4 5 2 5 1 2 5 4 2 2 4 7 4 1 Table 2: Statistics for various corpora utilized in experiments.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"From this we see that a finer-grained model for parser combination, at least for the features we have examined, will not give us any additional power.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Within this framework, we use features intended to capture degree of generality, including the output from an SVM classifier that uses the intersection between IN and OUT as positive examples.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"lla\/llb and 14a\/14b respectively).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Thus in an English sentence such as I'm going to show up at the ACL one would reasonably conjecture that there are eight words separated by seven spaces.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We allow any number of bits from 2 to 25, unlike IRSTLM (8 bits) and BerkeleyLM (17\u221220 bits).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The scoping heuristics are based on the anaphor type: for reflexive pronouns the scope is the current clause, for relative pronouns it is the prior clause following its VP, for personal pronouns it is the anaphor\u00e2\u0080\u0099s sentence and two preceding sentences, and for definite NPs it is the anaphor\u00e2\u0080\u0099s sentence and eight preceding sentences.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Nicola Bertoldi and Marcello Federico assisted with IRSTLM.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Besides information structure, the second main goal is to enhance current models of rhetorical structure.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"By using the output of a hand-coded system such as Proteus, MENE can improve its performance, and can even outperform IdentiFinder (Borthwick, 1999).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"In other words, the set of hidden states F was chosen to be the fine set of treebank tags.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A high-level relation is agent, which relates an animate nominal to a predicate.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Such resources exist for Hebrew (Itai et al., 2006), but unfortunately use a tagging scheme which is incompatible with the one of the Hebrew Treebank.s For this reason, we use a data-driven morphological analyzer derived from the training data similar to (Cohen and Smith, 2007).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For a word whose initCaps might be due to its position rather than its meaning (in headlines, first word of a sentence, etc), the case information of other occurrences might be more accurate than its own.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"A dynamic programming algorithm is then used to select the sequence of word classes with the highest probability.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"This left 962 examples, of which 85 were noise.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The rest of the paper is structured as follows.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A name already mentioned previously in a document may appear in abbreviated form when it is mentioned again later.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Intuitively, as suggested by the example in the introduction, this is the right granularity to capture domain effects.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For example, the passive voice pattern \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d and the active voice pattern \u00e2\u0080\u009ckidnapped <direct object>\u00e2\u0080\u009d are merged into a single normalized pattern \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d.2 For the sake of sim plicity, we will refer to these normalized extraction patterns as caseframes.3 These caseframes can capture two types of contextual role information: (1) thematic roles corresponding to events (e.g, \u00e2\u0080\u009c<agent> kidnapped\u00e2\u0080\u009d or \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d), and (2) predicate-argument relations associated with both verbs and nouns (e.g., \u00e2\u0080\u009ckidnapped for <np>\u00e2\u0080\u009d or \u00e2\u0080\u009cvehicle with <np>\u00e2\u0080\u009d).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"When extracting the vector t, used to compute the constraint feature from the graph, we tried three threshold values for r (see Eq.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Tables 4 and 5 show BABAR\u00e2\u0080\u0099s performance when just one contextual role knowledge source is used at a time.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"was done by the participants.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We excluded these from the evaluation as they can be easily identified with a list of days\/months.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"\u00e2\u0088\u0097NP NP PP R) and \u00e2\u0088\u0097NP NP ADJP R) are both iDafa attachment.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"If the context wnf will never extend to the right (i.e. wnf v is not present in the model for all words v) then no subsequent query will match the full context.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"For instance, on Spanish, the absolute gap on median performance is 10%.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"For these models we limit the options provided for OOV words by not considering the entire token as a valid segmentation in case at least some prefix segmentation exists.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"While the semantic aspect of radicals is by no means completely predictive, the semantic homogeneity of many classes is quite striking: for example 254 out of the 263 examples (97%) of the INSECT class listed by Wieger (1965, 77376) denote crawling or invertebrate animals; similarly 21 out of the 22 examples (95%) of the GHOST class (page 808) denote ghosts or spirits.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"It is striking that from this point of view many formalisms can be grouped together as having identically structured derivation tree sets.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"(Again, the goal of also in structural features.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Among these are words derived by various productive processes, including: 1.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"This annotation choice weakens splitIN.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"None of the models we have presented utilize features associated with a particular constituent (i.e. the label, span, parent label, etc.) to influence parser preference.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"1 The apparent difficulty of adapting constituency models to non-configurational languages has been one motivation for dependency representations (Hajic\u00cb\u0087 and Zema\u00c2\u00b4nek, 2004; Habash and Roth, 2009).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Training under this model involves estimation of parameter values for P(y), P(m) and P(x I y).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Every parse \u03c0 selects a specific morphological segmentation (l1...lk) (a path through the lattice).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"basically complete, yet some improvements and extensions are still under way.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The edge from the root to the subtree for the derivation of 7i is labeled by the address ni.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"But the city name Sharm Al- Sheikh is also iDafa, hence the possibility for the incorrect annotation in (b).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Now assume we have n pairs (xi,, x2,i) drawn from X1 X X2, where the first m pairs have labels whereas for i = m+ 1...n the pairs are unlabeled.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Instead, we condition on the type-level tag assignments T . Specifically, let St = {i|Ti = t} denote the indices of theword types which have been assigned tag t accord ing to the tag assignments T . Then \u00ce\u00b8t is drawn from DIRICHLET(\u00ce\u00b1, St), a symmetric Dirichlet which only places mass on word types indicated by St. This ensures that each word will only be assigned a single tag at inference time (see Section 4).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Unsupervised Learning of Contextual Role Knowledge for Coreference Resolution","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"State is implemented in their scrolling variant, which is a trie annotated with forward and backward pointers.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The first thing to note is that projectivizing helps in itself, even if no encoding is used, as seen from the fact that the projective baseline outperforms the non-projective training condition by more than half a percentage point on attachment score, although the gain is much smaller with respect to exact match.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"While there are other obstacles to completing this idea, we believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","label":1},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Note also that the costs currently used in the system are actually string costs, rather than word costs.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"By contrast, BerkeleyLM\u2019s hash and compressed variants will return incorrect results based on an n \u22121-gram.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Table 2 shows results for both settings and all methods described in sections 2 and 3.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"German city names are replaced by category markers.","label":0},{"source_text":"Here we present two algorithms.","target_text":"123 examples fell into the noise category.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"If they are found in a list, then a feature for that list will be set to 1.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For each resolution in the training data, BABAR also associates the co-referring expression of an NP with the NP\u00e2\u0080\u0099s caseframe.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"3.2 Stochastic rhetorical analysis.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"\u00e2\u0080\u00a2 Anaphoric links: the annotator is asked to specify whether the anaphor is a repetition, partial repetition, pronoun, epithet (e.g., Andy Warhol \u00e2\u0080\u0093 the PopArt artist), or is-a (e.g., Andy Warhol was often hunted by photographers.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"German city names are replaced by category markers.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"splitIN captures the verb\/preposition idioms that are widespread in Arabic.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The edge weights between the foreign language trigrams are computed using a co-occurence based similarity function, designed to indicate how syntactically similar the middle words of the connected trigrams are (\u00a73.2).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"These make left-to-right query patterns convenient, as the application need only provide a state and the word to append, then use the returned state to append another word, etc.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The points enumerated above are particularly related to ITS, but analogous arguments can easily be given for other applications; see for example Wu and Tseng's (1993) discussion of the role of segmentation in information retrieval.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Our open-source (LGPL) implementation is also available for download as a standalone package with minimal (POSIX and g++) dependencies.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Section 3 discusses the applications that have been completed with PCC, or are under way, or are planned for the future.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"For \u00c3\u0086 = 1, a new target language word is generated using the trigram language model p(eje0; e00).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Taking the intersection of languages in these resources, and selecting languages with large amounts of parallel data, yields the following set of eight Indo-European languages: Danish, Dutch, German, Greek, Italian, Portuguese, Spanish and Swedish.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"32 81.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For RandLM and IRSTLM, the effect of caching can be seen on speed and memory usage.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Despite these limitations, a purely finite-state approach to Chinese word segmentation enjoys a number of strong advantages.","label":0},{"source_text":"There is no global pruning.","target_text":"The algorithm works due to the fact that not all permutations of cities have to be considered explicitly.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"(we ignored the 419 trees in their development set.)","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"For example, the MCTAG shown in Figure 7 generates trees of the form shown in Figure 4b.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Almost all annotators expressed their preference to move to a ranking-based evaluation in the future.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"For example, one of the ATB samples was the determiner -\"\" ; dhalik\u00e2\u0080\u009cthat.\u00e2\u0080\u009d The sample occurred in 1507 corpus po sitions, and we found that the annotations were consistent.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The set of n-grams appearing in a model is sparse, and we want to efficiently find their associated probabilities and backoff penalties.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Specifically, for the ith word type, the set of token-level tags associated with token occurrences of this word, denoted t(i), must all take the value Ti to have nonzero mass. Thus in the context of Gibbs sampling, if we want to block sample Ti with t(i), we only need sample values for Ti and consider this setting of t(i).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Here, an NE instance pair is any pair of NEs separated by at most 4 syntactic chunks; for example, \u00e2\u0080\u009cIBM plans to acquire Lotus\u00e2\u0080\u009d.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"43 80.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"An example of a fairly low-level relation is the affix relation, which holds between a stem morpheme and an affix morpheme, such as f1 -menD (PL).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"4.1 The Task and the Corpus.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"This vector tx is constructed for every word in the foreign vocabulary and will be used to provide features for the unsupervised foreign language POS tagger.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"IRST is not threadsafe.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The final model tions.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"However, to classify a token , while Borthwick uses tokens from to (from two tokens before to two tokens after ), we used only the tokens , , and . Even with local features alone, MENERGI outperforms MENE (Borthwick, 1999).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"AdaBoost.MH maintains a distribution over instances and labels; in addition, each weak-hypothesis outputs a confidence vector with one confidence value for each possible label.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"It is difficult to directly compare the Matsoukas et al results with ours, since our out-of-domain corpus is homogeneous; given heterogeneous training data, however, it would be trivial to include Matsoukas-style identity features in our instance-weighting model.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Two sets of examples from Gan are given in (1) and (2) (:::::: Gan's Appendix B, exx.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The features are weighted within a logistic model to give an overall weight that is applied to the phrase pair\u2019s frequency prior to making MAP-smoothed relative-frequency estimates (different weights are learned for each conditioning direction).","label":1},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"When extracting the vector t, used to compute the constraint feature from the graph, we tried three threshold values for r (see Eq.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"When this feature type was included, CoBoost chose this default feature at an early iteration, thereby giving non-abstaining pseudo-labels for all examples, with eventual convergence to the two classifiers agreeing by assigning the same label to almost all examples.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In Input: {(x1,i, Initialize: Vi, j : e(xi) = 0.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2009).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"In order to evaluate and advance this approach, it helps to feed into the knowledge base data that is already enriched with some of the desired information \u00e2\u0080\u0094 as in PCC.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"3.2 Stochastic rhetorical analysis.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The only way to handle such phenomena within the framework described here is simply to expand out the reduplicated forms beforehand, and incorporate the expanded forms into the lexical transducer.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"But we will show that the use of unlabeled data can drastically reduce the need for supervision.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Thus, provided at least this amount of IN data is available\u2014as it is in our setting\u2014adapting these weights is straightforward.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"TPT has theoretically better locality because it stores ngrams near their suffixes, thereby placing reads for a single query in the same or adjacent pages.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"One key issue here is to seek a discourse-based model of information structure.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Context from the whole document can be important in classifying a named entity.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Realizing gains in practice can be challenging, however, particularly when the target domain is distant from the background data.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Riloff and Jones 99) was brought to our attention as we were preparing the final version of this paper.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Later, BerkeleyLM (Pauls and Klein, 2011) described ideas similar to ours.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We do not show the numbers for the Bayes models in Table 2 because the parameters involved were established using this set.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Word frequencies are estimated by a re-estimation procedure that involves apply\u00c2\u00ad ing the segmentation algorithm presented here to a corpus of 20 million words,8 using 8 Our training corpus was drawn from a larger corpus of mixed-genre text consisting mostly of.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The constituent voting and na\u00efve Bayes techniques are equivalent because the parameters learned in the training set did not sufficiently discriminate between the three parsers.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"To optimize this function, we used L-BFGS, a quasi-Newton method (Liu and Nocedal, 1989).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"This can be seen as a rough approximation of Yarowsky and Ngai (2001).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For fair comparison, we have tabulated all results with the size of training data used (Table 5 and Table 6).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"An input ABCD can be represented as an FSA as shown in Figure 2(b).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"For each language under consideration, Petrov et al. (2011) provide a mapping A from the fine-grained language specific POS tags in the foreign treebank to the universal POS tags.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Step 2.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"However, if we consider precision, recall and Fmeasure on non-projective dependencies only, as shown in Table 6, some differences begin to emerge.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"La ng ua ge # To ke ns # W or d Ty pe s # Ta gs E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 1 1 7 3 7 6 6 9 4 3 8 6 2 0 3 5 6 8 6 9 9 6 0 5 2 0 6 6 7 8 8 9 3 3 4 1 9 1 4 6 7 4 9 2 0 6 1 8 3 5 6 2 8 3 9 3 7 2 3 2 5 2 8 9 3 1 1 6 4 5 8 2 0 0 5 7 4 5 2 5 1 2 5 4 2 2 4 7 4 1 Table 2: Statistics for various corpora utilized in experiments.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The first issue relates to the completeness of the base lexicon.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"If one is interested in translation, one would probably want to consider show up as a single dictionary word since its semantic interpretation is not trivially derivable from the meanings of show and up.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We tagged each noun with the top-level semantic classes assigned to it in Word- Net.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Still, for both human and automatic rhetorical analysis, connectives are the most important source of surface information.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"First we mark any node that dominates (at any level) a verb sider POS tags when pre-terminals are the only intervening nodes between the nucleus and its bracketing (e.g., unaries, base NPs).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(3)) to be defined over unlabeled as well as labeled instances.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"These sequence models-based approaches commonly treat token-level tag assignment as the primary latent variable.","label":1},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The final estimating equation is then: (3) Since the total of all these class estimates was about 10% off from the Turing estimate n1\/N for the probability of all unseen hanzi, we renormalized the estimates so that they would sum to n 1jN.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"In an experiment on automatic rhetorical parsing, the RST-annotations and PoS tags were used by (Reitter 2003) as a training corpus for statistical classification with Support Vector Machines.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Before explaining our method in detail, we present a brief overview in this subsection.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We evaluate the time and memory consumption of each data structure by computing perplexity on 4 billion tokens from the English Gigaword corpus (Parker et al., 2009).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"However, our full model takes advantage of word features not present in Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"While the semantic aspect of radicals is by no means completely predictive, the semantic homogeneity of many classes is quite striking: for example 254 out of the 263 examples (97%) of the INSECT class listed by Wieger (1965, 77376) denote crawling or invertebrate animals; similarly 21 out of the 22 examples (95%) of the GHOST class (page 808) denote ghosts or spirits.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"level.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"By this assumption, each element x E X can also be represented as (xi, x2) E X1 x X2.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"2.3 Assigning Evidence Values.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"For parsing, the most challenging form of ambiguity occurs at the discourse level.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"00 76.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"However, when grammatical relations like subject and object are evaluated, parsing performance drops considerably (Green et al., 2009).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, until such standards are universally adopted in evaluating Chinese segmenters, claims about performance in terms of simple measures like percent correct should be taken with a grain of salt; see, again, Wu and Fung (1994) for further arguments supporting this conclusion.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Let 71 be a tree with root and foot labeled by X.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"If either case is true, then CFLex reports that the anaphor and candidate might be coreferent.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"(Levinger et al., 1995; Goldberg et al., ; Adler et al., 2008)) will make the parser more robust and suitable for use in more realistic scenarios.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, if CFLex determines that the log- likelihood statistic for the co-occurrence of a particular noun and caseframe corresponds to the 90% confidence level, then CFLex returns .90 as its belief that the anaphor and candidate are coreferent.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"There has been additional recent work on inducing lexicons or other knowledge sources from large corpora.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"On MUC6 data, MENERGI also achieves performance comparable to IdentiFinder when trained on similar amount of training data.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Figure 2 shows an excerpt of a sentence from the Italian test set and the tags assigned by four different models, as well as the gold tags.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"This model admits a simple Gibbs sampling algorithm where the number of latent variables is proportional to the number of word types, rather than the size of a corpus as for a standard HMM sampler (Johnson, 2007).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Corporate Suffixes and Person Prefixes of Other Occurrences (CSPP): If McCann has been seen as Mr. McCann somewhere else in the document, then one would like to give person a higher probability than organization.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Since MUC6, BBN' s Hidden Markov Model (HMM) based IdentiFinder (Bikel et al., 1997) has achieved remarkably good performance.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"With the same Corporate- Suffix-List and Person-Prefix-List used in local features, for a token seen elsewhere in the same document with one of these suffixes (or prefixes), another feature Other-CS (or Other-PP) is set to 1.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In figure 4, reverse relations are indicated by `*\u00e2\u0080\u0099 next to the frequency.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We group the features used into feature groups.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"RandLM and SRILM also remove context that will not extend, but SRILM performs a second lookup in its trie whereas our approach has minimal additional cost.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"This model admits a simple Gibbs sampling algorithm where the number of latent variables is proportional to the number of word types, rather than the size of a corpus as for a standard HMM sampler (Johnson, 2007).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"For instance, by altering the emission distribution parameters, Johnson (2007) encourages the model to put most of the probability mass on few tags.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Limitations of (Blum and Mitchell 98): While the assumptions of (Blum and Mitchell 98) are useful in developing both theoretical results and an intuition for the problem, the assumptions are quite limited.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For each domain, we created a semantic dictionary by doing two things.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"On each language we investigate the contribution of each component of our model.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The relativizer f(\u201cthat\u201d) for example, may attach to an arbitrarily long relative clause that goes beyond token boundaries.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"As expected, the vanilla HMM trained with EM performs the worst.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"We can do that . IbmS: Yes, wonderful.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The first row represents the average accuracy of the three parsers we combine.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"However their work did not consider other types of lexical expectations (e.g., PP arguments), semantic expectations, or context comparisons like our case- frame network.(Niyu et al., 1998) used unsupervised learning to ac quire gender, number, and animacy information from resolutions produced by a statistical pronoun resolver.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"BerkeleyLM uses states to optimistically search for longer n-gram matches first and must perform twice as many random accesses to retrieve backoff information.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If is one of Monday, Tuesday, . . .","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Conditioned on T , features of word types W are drawn.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The counts represent portions of the approximately 44000 constituents hypothesized by the parsers in the development set.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For example, in predicting if a word belongs to a word class, is either true or false, and refers to the surrounding context: if = true, previous word = the otherwise The parameters are estimated by a procedure called Generalized Iterative Scaling (GIS) (Darroch and Ratcliff, 1972).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The basic strategy is, for a given pair of entity types, to start with some examples, like several famous book title and author pairs; and find expressions which contains those names; then using the found expressions, find more author and book title pairs.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"When using the segmentation pruning (using HSPELL) for unseen tokens, performance improves for all tasks as well.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"5 60.6 Table 3: Multilingual Results: We report token-level one-to-one and many-to-one accuracy on a variety of languages under several experimental settings (Section 5).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"One is that smaller sets sometime have meaningless keywords, like \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d in the CC-domain, or \u00e2\u0080\u009ccompare\u00e2\u0080\u009d in the PC-domain.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Section 3 describes the complete coreference resolution model, which uses the contextual role knowledge as well as more traditional coreference features.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"On the other hand, in a translation system one probably wants to treat this string as a single dictionary word since it has a conventional and somewhat unpredictable translation into English.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We offer a state function s(wn1) = wn\ufffd where substring wn\ufffd is guaranteed to extend (to the right) in the same way that wn1 does for purposes of language modeling.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Furthermore, we do not connect the English vertices to each other, but only to foreign language vertices.4 The graph vertices are extracted from the different sides of a parallel corpus (De, Df) and an additional unlabeled monolingual foreign corpus Ff, which will be used later for training.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":".","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Under this scheme, n human judges are asked independently to segment a text.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2009) on Portuguese (Grac\u00c2\u00b8a et al.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The logistic function, whose outputs are in [0, 1], forces pp(s, t) <_ po(s, t).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Clearly it is possible to write a rule that states that if an analysis Modal+ Verb is available, then that is to be preferred over Noun+ Verb: such a rule could be stated in terms of (finite-state) local grammars in the sense of Mohri (1993).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"It is sometimes claimed that one of the advantages of dependency grammar over approaches based on constituency is that it allows a more adequate treatment of languages with variable word order, where discontinuous syntactic constructions are more common than in languages like English (Mel\u2019\u02c7cuk, 1988; Covington, 1990).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The Bayes models were able to achieve significantly higher precision than their non-parametric counterparts.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We use v1.0 mainly because previous studies on joint inference reported results w.r.t. v1.0 only.5 We expect that using the same setup on v2.0 will allow a crosstreebank comparison.6 We used the first 500 sentences as our dev set and the rest 4500 for training and report our main results on this split.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The links can solve the problem.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Finally, Section 5 explains how BABAR relates to previous work, and Section 6 summarizes our conclusions.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In this paper, we describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The P (T |\u00cf\u0088) distribution, in English for instance, should have very low mass for the DT (determiner) tag, since determiners are a very small portion of the vocabulary.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Both authors are members of the Center for Language and Speech Processing at Johns Hopkins University.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Extract NE pair instances with contexts From the four years of newspaper corpus, we extracted 1.9 million pairs of NE instances.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"These make left-to-right query patterns convenient, as the application need only provide a state and the word to append, then use the returned state to append another word, etc.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"2.6 Co-reference.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Methods for expanding the dictionary include, of course, morphological rules, rules for segmenting personal names, as well as numeral sequences, expressions for dates, and so forth (Chen and Liu 1992; Wang, Li, and Chang 1992; Chang and Chen 1993; Nie, Jin, and Hannan 1994).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"They first collect the NE instance pairs and contexts, just like our method.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Arabic sentences of up to length 63 would need to be.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Thus, the derivation trees for TAG's have the same structure as local sets.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The motivation for our more informal approach was the intuition that there are so many open problems in rhetorical analysis (and more so for German than for English; see below) that the main task is qualitative investigation, whereas rigorous quantitative analyses should be performed at a later stage.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"On the English side, however, the vertices (denoted by Ve) correspond to word types.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Naseem et al. (2009) and Snyder et al.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"We have tested the translation system on the Verbmobil task (Wahlster 1993).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"But we follow the more direct adaptation of Evalb suggested by Tsarfaty (2006), who viewed exact segmentation as the ultimate goal.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For a description of the application of AdaBoost to various NLP problems see the paper by Abney, Schapire, and Singer in this volume.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In the second and third translation examples, the IbmS word reordering performs worse than the QmS word reordering, since it can not take properly into account the word reordering due to the German verbgroup.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The family name set is restricted: there are a few hundred single-hanzi family names, and about ten double-hanzi ones.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The index in this array is the vocabulary identifier.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Finally, the assignments of PoS tags to OOV segments is subject to language specific constraints relative to the token it was originated from.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If is a number string (such as one, two, etc), then the feature NumberString is set to 1.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"There are 13 types of extensions needed to describe the verbgroup reordering.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Finally, we intend to explore more sophisticated instanceweighting features for capturing the degree of generality of phrase pairs.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"2.4 Underspecified rhetorical structure.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"2.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Our oracles took advantage of the labeled treebanks: While we tried to minimize the number of free parameters in our model, there are a few hyperparameters that need to be set.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Figure 4 shows the seven general knowledge sources (KSs) that represent features commonly used for coreference resolution.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Exposing this information to the decoder will lead to better hypothesis recombination.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"It is generally considered to be fast (Pauls 29 \u2212 1 probabilities and 2' \u2212 2 non-zero backoffs. and Klein, 2011), with a default implementation based on hash tables within each trie node.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The English data comes from the WSJ portion of the Penn Treebank and the other languages from the training set of the CoNLL-X multilingual dependency parsing shared task.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If is seen infrequently during training (less than a small count), then will not be selected as a feature and all features in this group are set to 0.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"More examples are shown in Figure 5.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Our TRIE implementation is designed to improve upon IRSTLM using a reverse trie with improved search, bit level packing, and stateful queries.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"1).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Several coreference resolvers have used supervised learning techniques, such as decision trees and rule learners (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":1},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"2 60.7 56.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Section 2.1 describes how BABAR generates training examples to use in the learning process.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"However, there are phrases which express the same meanings even though they do not share the same keyword.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Interpolation search is therefore a form of binary search with better estimates informed by the uniform key distribution.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"yields.6 But to its great advantage, it has a high ratio of non-terminals\/terminals (\u00ce\u00bc Constituents \/ \u00ce\u00bc Length).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The interdependence between fb or 1\/!i, and 5:2 is not captured by our model, but this could easily be remedied.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Specifically, we (+FEATS) utilizes the tag prior as well as features (e.g., suffixes and orthographic features), discussed in Section 3, for the P (W |T , \u00cf\u0088) component.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Compared to related work on the recovery of long-distance dependencies in constituency-based parsing, our approach is similar to that of Dienes and Dubey (2003) in that the processing of non-local dependencies is partly integrated in the parsing process, via an extension of the set of syntactic categories, whereas most other approaches rely on postprocessing only.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Also there are cases where one of the two NEs belong to a phrase outside of the relation.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"After adding a ROOT node to all trees, we train a grammar using six split-and- merge cycles and no Markovization.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"There are thus some very good reasons why segmentation into words is an important task.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"9 61.0 44.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The maximum precision oracle is an upper bound on the possible gain we can achieve by parse hybridization.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Previous reports on Chinese segmentation have invariably cited performance either in terms of a single percent-correct score, or else a single precision-recall pair.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Another question that remains unanswered is to what extent the linguistic information he considers can be handled-or at least approximated-by finite-state language models, and therefore could be directly interfaced with the segmentation model that we have presented in this paper.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In addition to the named-entity string (Maury Cooper or Georgia), a contextual predictor was also extracted.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"For illustration, an English translation of one of the commentaries is given in Figure 1.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The SynRole KS computes the relative frequency with which the candidates\u00e2\u0080\u0099 syntactic role (subject, direct object, PP object) appeared in resolutions in the training set.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"This algorithm can be applied to statistical machine translation.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"These three parsers have given the best reported parsing results on the Penn Treebank Wall Street Journal corpus (Marcus et al., 1993).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"For instance, the sentence Similar improvements in haemoglobin levels were reported in the scientific literature for other epoetins would likely be considered domain-specific despite the presence of general phrases like were reported in.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The tokens w are generated by token-level tags t from an HMM parameterized by the lexicon structure.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Furthermore, we do not connect the English vertices to each other, but only to foreign language vertices.4 The graph vertices are extracted from the different sides of a parallel corpus (De, Df) and an additional unlabeled monolingual foreign corpus Ff, which will be used later for training.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"(b) POS tagging accuracy is lowest for maSdar verbal nouns (VBG,VN) and adjectives (e.g., JJ).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Table 1 The cost as a novel given name (second position) for hanzi from various radical classes.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"SRILM (Stolcke, 2002) is widely used within academia.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The advantage is that we can recombine search hypotheses by dynamic programming.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"att.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"4.3 Morphological Analysis.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"3 The Coreference Resolution Model.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"For a trigram language model, the partial hypotheses are of the form (e0; e; C; j).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Arabic sentences of up to length 63 would need to be.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We have developed a coreference resolver called BABAR that uses contextual role knowledge to make coreference decisions.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"To combat the proliferation of parsing edges, we prune the lattices according to a hand-constructed lexicon of 31 clitics listed in the ATB annotation guidelines (Maamouri et al., 2009a).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"RandLM and SRILM also remove context that will not extend, but SRILM performs a second lookup in its trie whereas our approach has minimal additional cost.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Even this may be nondeterministic, in case the graph contains several non-projective arcs whose lifts interact, but we use the following algorithm to construct a minimal projective transformation D0 = (W, A0) of a (nonprojective) dependency graph D = (W, A): The function SMALLEST-NONP-ARC returns the non-projective arc with the shortest distance from head to dependent (breaking ties from left to right).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"7 Conclusion and Future Work.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, it is almost universally the case that no clear definition of what constitutes a \"correct\" segmentation is given, so these performance measures are hard to evaluate.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"To this end, we picked 100 sentences at random containing 4,372 total hanzi from a test corpus.14 (There were 487 marks of punctuation in the test sentences, including the sentence-final periods, meaning that the average inter-punctuation distance was about 9 hanzi.)","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"If the semantic expectations do not intersect, then we know that the case- frames extract mutually exclusive types of noun phrases.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The method uses a &quot;soft&quot; measure of the agreement between two classifiers as an objective function; we described an algorithm which directly optimizes this function.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"First, we identify sources of syntactic ambiguity understudied in the existing parsing literature.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"A generative model was applied (similar to naive Bayes) with the three labels as hidden vanables on unlabeled examples, and observed variables on (seed) labeled examples.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Fortunately, we were able to obtain a copy of the full set of sentences from Chang et al. on which Wang, Li, and Chang tested their system, along with the output of their system.18 In what follows we will discuss all cases from this set where our performance on names differs from that of Wang, Li, and Chang.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For example, given a sequence F1G1G2, where F1 is a legal single-hanzi family name, and Plural Nouns X g 0 g \"' X X 0 T!i c\"'.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Note, however, that there might be situations in which Zco in fact increases.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Figure 3 Figure 1.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"(2) Description Source Location Names http:\/\/www.timeanddate.com http:\/\/www.cityguide.travel-guides.com http:\/\/www.worldtravelguide.net Corporate Names http:\/\/www.fmlx.com Person First Names http:\/\/www.census.gov\/genealogy\/names Person Last Names Table 2: Sources of Dictionaries The McCann family . . .","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"12 For English, our Evalb implementation is identical to the most recent reference (EVALB20080701).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"For example, one of the ATB samples was the determiner -\"\" ; dhalik\u00e2\u0080\u009cthat.\u00e2\u0080\u009d The sample occurred in 1507 corpus po sitions, and we found that the annotations were consistent.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The developer explained that the loading process requires extra memory that it then frees. eBased on the ratio to SRI\u2019s speed reported in Guthrie and Hepple (2010) under different conditions.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Though we do not directly compare state implementations, performance metrics in Table 1 indicate our overall method is faster.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We excluded these from the evaluation as they can be easily identified with a list of days\/months.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"P St = n. \u00ce\u00b2 T VARIABLES \u00cf\u0088 Y W : Word types (W1 ,.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The focus of our work is on the use of contextual role knowledge for coreference resolution.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"segmentation (Table 2).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"An important aspect of the DempsterShafer model is that it operates on sets of hypotheses.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In Figure 4 we show an example of variation between the parsing models.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Instead, the designs of the various annotation layers and the actual annotation work are results of a series of diploma theses, of students\u00e2\u0080\u0099 work in course projects, and to some extent of paid assistentships.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Note that Chang, Chen, and Chen (1991), in addition to word-frequency information, include a constraint-satisfication model, so their method is really a hybrid approach.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"As shown in Table 3, the proportion of sentences containing some non-projective dependency ranges from about 15% in DDT to almost 25% in PDT.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"36.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Yarowsky-cautious does not separate the spelling and contextual features, but does have a limit on the number of rules added at each stage.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"More recently, Subramanya et al. (2010) defined a graph over the cliques in an underlying structured prediction model.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For example, even if the contexts surrounding an anaphor and candidate match exactly, they are not coreferent if they have substantially different meanings 9 We would be happy to make our manually annotated test data available to others who also want to evaluate their coreference resolver on the MUC4 or Reuters collections.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":" We evaluate the system's performance by comparing its segmentation 'Tudgments\" with the judgments of a pool of human segmenters, and the system is shown to perform quite well.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"na me =>1 ha nzi fa mi ly 2 ha nzi gi ve n 3.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Given parameter estimates, the label for a test example x is defined as We should note that the model in equation 9 is deficient, in that it assigns greater than zero probability to some feature combinations that are impossible.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"As noted in Section 4.4, disk cache state is controlled by reading the entire binary file before each test begins.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"This can be seen as a rough approximation of Yarowsky and Ngai (2001).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"5 68.1 34.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Only IRSTLM does not support threading.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"5 60.6 Table 3: Multilingual Results: We report token-level one-to-one and many-to-one accuracy on a variety of languages under several experimental settings (Section 5).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"After the first step towards breadth had been taken with the PoS-tagging, RST annotation, and URML conversion of the entire corpus of 170 texts12 , emphasis shifted towards depth.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"We can do that . Input: Das ist zu knapp , weil ich ab dem dritten in Kaiserslautern bin . Genaugenommen nur am dritten . Wie ware es denn am ahm Samstag , dem zehnten Februar ? MonS: That is too tight , because I from the third in Kaiserslautern . In fact only on the third . How about ahm Saturday , the tenth of February ? QmS: That is too tight , because I am from the third in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? IbmS: That is too tight , from the third because I will be in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? Input: Wenn Sie dann noch den siebzehnten konnten , ware das toll , ja . MonS: If you then also the seventeenth could , would be the great , yes . QmS: If you could then also the seventeenth , that would be great , yes . IbmS: Then if you could even take seventeenth , that would be great , yes . Input: Ja , das kommt mir sehr gelegen . Machen wir es dann am besten so . MonS: Yes , that suits me perfectly . Do we should best like that . QmS: Yes , that suits me fine . We do it like that then best . IbmS: Yes , that suits me fine . We should best do it like that .","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Our TRIE implements the popular reverse trie, in which the last word of an n-gram is looked up first, as do SRILM, IRSTLM\u2019s inverted variant, and BerkeleyLM except for the scrolling variant.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"This annotation choice weakens splitIN.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Given a document to process, BABAR uses four modules to perform coreference resolution.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The learned information was recycled back into the resolver to improve its performance.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Other errors include NE tagging errors and errors due to a phrase which includes other NEs.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Otherwise, it is set to 0.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In the terrorism domain, 1600 texts were used for training and the 40 test docu X \u00e2\u0088\u00a9Y =\u00e2\u0088\u0085 All sets of hypotheses (and their corresponding belief values) in the current model are crossed with the sets of hypotheses (and belief values) provided by the new evidence.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In order to create an IE system for a new domain, one has to spend a long time to create the knowledge.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This flexibility, along with the simplicity of implementation and expansion, makes this framework an attractive base for continued research.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The first 3770 trees of the resulting set then were used for training, and the last 418 are used testing.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In this paper, we will propose an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The search starts in the hypothesis (I; f;g; 0).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"These words are in turn highly ambiguous, breaking the assumption underlying most parsers that the yield of a tree for a given sentence is known in advance.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Oracle results).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Of course, we.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Somewhat surprisingly, there do not appear to be large systematic differences between linear and MAP combinations.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"On the surface, our model may seem as a special case of Cohen and Smith in which \u03b1 = 0.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"3.2 Reordering with IBM Style.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"However there is no global pruning.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"None of the models attach the attributive adjectives correctly.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The terms on the right-hand-side denote the type-level and token-level probability terms respectively.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We thank United Informatics for providing us with our corpus of Chinese text, and BDC for the 'Behavior ChineseEnglish Electronic Dictionary.'","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(Blum and Mitchell 98) describe learning in the following situation: X = X1 X X2 where X1 and X2 correspond to two different &quot;views&quot; of an example.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"If a phrase does not contain any keywords, the phrase is discarded.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Coverage indicates the fraction of hypotheses in which the character yield exactly matched the reference.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We tried two versions of our graph-based approach: feature after the first stage of label propagation (Eq.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We have described grammar state splits that significantly improve parsing performance, catalogued parsing errors, and quantified the effect of segmentation errors.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Each spawned process must check if xi , , xn, and , yn, can be derived from B and C, respectively.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"This is the parse that is closest to the centroid of the observed parses under the similarity metric.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We can only compare with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We are not claiming that this method is almighty.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"One of the difficulties in Natural Language Processing is the fact that there are many ways to express the same thing or event.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The set of candidate constituents comes from the union of all the constituents suggested by the member parsers.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(a) 1 \u00c2\u00a7 . ;m t 7 leO z h e 4 pil m a 3 lu 4 sh an g4 bi ng 4 t h i s CL (assi fier) horse w ay on sic k A SP (ec t) 'This horse got sick on the way' (b) 1\u00c2\u00a7: . til y zhe4 tiao2 ma3lu4 hen3 shao3 this CL road very few 'Very few cars pass by this road' :$ chel jinglguo4 car pass by 2.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The hash variant is a reverse trie with hash tables, a more memory-efficient version of SRILM\u2019s default.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In the second scheme, Head+Path, we in addition modify the label of every arc along the lifting path from the syntactic to the linear head so that if the original label is p the new label is p\u2193.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Unsupervised Part-of-Speech Tagging with Bilingual Graph-Based Projections","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Our model outperforms theirs on four out of five languages on the best hyperparameter setting and three out of five on the median setting, yielding an average absolute difference across languages of 12.9% and 3.9% for best and median settings respectively compared to their best EM or LBFGS performance.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"na me =>2 ha nzi fa mi ly 2 ha nzi gi ve n 5.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Overview of the method 2.2 Step by Step Algorithm.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The web-based Annis imports data in a variety of XML formats and tagsets and displays it in a tier-orientedway (optionally, trees can be drawn more ele gantly in a separate window).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In this section, we try to compare our results with those obtained by IdentiFinder ' 97 (Bikel et al., 1997), IdentiFinder ' 99 (Bikel et al., 1999), and MENE (Borthwick, 1999).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Now assume we have n pairs (xi,, x2,i) drawn from X1 X X2, where the first m pairs have labels whereas for i = m+ 1...n the pairs are unlabeled.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"There is no relation between these two interpretations other then the fact that their surface forms coincide, and we argue that the only reason to prefer one analysis over the other is compositional.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Whether a language even has orthographic words is largely dependent on the writing system used to represent the language (rather than the language itself); the notion \"orthographic word\" is not universal.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"This limitation is the obstacle to making the technology \u00e2\u0080\u009copen domain\u00e2\u0080\u009d.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"An interesting observation is that while vertical markovization benefits all our models, its effect is less evident in Cohen and Smith.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Trees derived by IG's exhibit a property that is not exhibited by the trees sets derived by TAG's or CFG's.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The development set contained 44088 constituents in 2416 sentences and the test set contained 30691 constituents in 1699 sentences.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"These methods demonstrated the benefits of incorporating linguistic features using a log-linear parameterization, but requires elaborate machinery for training.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"It was developed in response to the non-terminal\/terminal bias of Evalb, but Clegg and Shepherd (2005) showed that it is also a valuable diagnostic tool for trees with complex deep structures such as those found in the ATB.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"This approach needs a phrase as an initial seed and thus the possible relationships to be extracted are naturally limited.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"It is natural that the larger the data in the domain, the more keywords are found.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In such cases we assign all of the estimated probability mass to the form with the most likely pronunciation (determined by inspection), and assign a very small probability (a very high cost, arbitrarily chosen to be 40) to all other variants.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"On the other hand, in a translation system one probably wants to treat this string as a single dictionary word since it has a conventional and somewhat unpredictable translation into English.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"At first glance, the problem seems quite complex: a large number of rules is needed to cover the domain, suggesting that a large number of labeled examples is required to train an accurate classifier.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Using the concept of inverted alignments, we explicitly take care of the coverage constraint by introducing a coverage set C of source sentence positions that have been already processed.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"At this stage the lattice path corresponds to segments only, with no PoS assigned to them.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This task measures how well each package performs in machine translation.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The first shows how constituent features and context do not help in deciding which parser to trust.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"As noted for the perplexity task, we do not expect cache to grow substantially with model size, so RandLM remains a low-memory option.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"One knowledge source, called WordSemCFSem, is analogous to CFLex: it checks whether the anaphor and candidate antecedent are substitutable for one another, but based on their semantic classes instead of the words themselves.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"8 Volume 1, English language, 19961997, Format version 1, correction level 0 An ap ho r T e r r o r i s m R e c Pr F D i s a s t e r s R e c Pr F De f. NP s Pro no uns .43 .79 .55 .50 .72 .59 .42 .91 .58 .42 .82 .56 Tot al .46 .76 .57 .42 .87 .57 Table 2: General Knowledge Sources Table 4: Individual Performance of KSs for Terrorism Table 3: General + Contextual Role Knowledge Sources larger MUC4 and Reuters corpora.9 4.2 Experiments.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"This result suggests the benefit of using the automatic discovery method.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"A novel element of our model is the ability to capture type-level tag frequencies.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Every token is independent of the others, and the sentence lattice is in fact a concatenation of smaller lattices, one for each token.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The SynRole KS computes the relative frequency with which the candidates\u00e2\u0080\u0099 syntactic role (subject, direct object, PP object) appeared in resolutions in the training set.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The following algorithm was then used to induce new rules: Let Count' (x) be the number of times feature x is seen with some known label in the training data.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Our system does not currently make use of titles, but it would be straightforward to do so within the finite-state framework that we propose.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"68 95.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Once we figure out the important word (e.g. keyword), we believe we can capture the meaning of the phrase by the keyword.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The best analysis of the corpus is taken to be the true analysis, the frequencies are re-estimated, and the algorithm is repeated until it converges.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(Blum and Mitchell 98) offer a promising formulation of redundancy, also prove some results about how the use of unlabeled examples can help classification, and suggest an objective function when training with unlabeled examples.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"If (wi, r, wj) E A, we say that wi is the head of wj and wj a dependent of wi.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Similarly, hanzi sharing the GHOST radical _m tend to denote spirits and demons, such as _m gui3 'ghost' itself, II: mo2 'demon,' and yan3 'nightmare.'","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"When two partial hypotheses have equal state (including that of other features), they can be recombined and thereafter efficiently handled as a single packed hypothesis.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Gather phrases using keywords Now, the keyword with the top TF\/ITF score is selected for each phrase.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"5 70.1 58.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"First, in section 4, we evaluate the graph transformation techniques in themselves, with data from the Prague Dependency Treebank and the Danish Dependency Treebank.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"But we will show that the use of unlabeled data can drastically reduce the need for supervision.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For example, BABAR learned that agents that \u00e2\u0080\u009cassassinate\u00e2\u0080\u009d or \u00e2\u0080\u009cinvestigate a cause\u00e2\u0080\u009d are usually humans or groups (i.e., organizations).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"So, who won the competition?","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Automatic Paraphrase Discovery based on Context and Keywords between NE Pairs","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"(2009) also report results on English, but on the reduced 17 tag set, which is not comparable to ours).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"However, for our purposes it is not sufficient to repre\u00c2\u00ad sent the morphological decomposition of, say, plural nouns: we also need an estimate of the cost of the resulting word.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"It can also be seen clearly in this plot that two of the Taiwan speakers cluster very closely together, and the third Tai\u00c2\u00ad wan speaker is also close in the most significant dimension (the x axis).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Finally, we incorporate the instance-weighting model into a general linear combination, and learn weights and mixing parameters simultaneously. where c\u03bb(s, t) is a modified count for pair (s, t) in OUT, u(s|t) is a prior distribution, and y is a prior weight.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Table 2 shows our complete set of results.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"If one of these checks fails then this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Cluster phrases based on Links We now have a set of phrases which share a keyword.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Clearly, explicitly modeling such a powerful constraint on tagging assignment has a potential to significantly improve the accuracy of an unsupervised part-of-speech tagger learned without a tagging dictionary.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"For example, in Figure 3, we can see that the phrases in the \u00e2\u0080\u009cbuy\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire\u00e2\u0080\u009d and \u00e2\u0080\u009cpurchase\u00e2\u0080\u009d sets are mostly paraphrases.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"(Carlson, Marcu 2001) responded to this situation with relatively precise (and therefore long!)","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"), which precludes a single universal approach to adaptation.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Among the IS-units, the referring expressions are marked as such and will in the second phase receive a label for cognitive status (active, accessible- text, accessible-situation, inferrable, inactive).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"3.1 Corpora.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"diesem 3.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For the \u00e2\u0080\u0098core\u00e2\u0080\u0099 portion of PCC, we found that on average, 35% of the coherence relations in our RST annotations are explicitly signalled by a lexical connective.6 When adding the fact that connectives are often ambiguous, one has to conclude that prospects for an automatic analysis of rhetorical structure using shallow methods (i.e., relying largely on connectives) are not bright \u00e2\u0080\u0094 but see Sections 3.2 and 3.3 below.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We train linear mixture models for conditional phrase pair probabilities over IN and OUT so as to maximize the likelihood of an empirical joint phrase-pair distribution extracted from a development set.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In this section for the purposes of showing that polynomial time recognition is possible, we make the additional restriction that the contribution of a derived structure to the input string can be specified by a bounded sequence of substrings of the input.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"This model admits a simple Gibbs sampling algorithm where the number of latent variables is proportional to the number of word types, rather than the size of a corpus as for a standard HMM sampler (Johnson, 2007).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Space- or punctuation-delimited * 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Hence, the terminal symbols appearing in the structures that are composed are not lost (though a constant number of new symbols may be introduced).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We are very grateful to Tony Kroc.h, Michael Pails, Sunil Shende, and Mark Steedman for valuable discussions. formalisms.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For every sequence of initial capitalized words, its longest substring that occurs in the same document as a sequence of initCaps is identified.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In principle a feature could be an arbitrary predicate of the (spelling, context) pair; for reasons that will become clear, features are limited to querying either the spelling or context alone.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"However, since we want to preserve as much of the original structure as possible, we are interested in finding a transformation that involves a minimal number of lifts.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Mikheev et al.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Sima\u2019an et al. (2001) presented parsing results for a DOP tree-gram model using a small data set (500 sentences) and semiautomatic morphological disambiguation.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For example, if Barry is not in commonWords and is found in the list of person first names, then the feature PersonFirstName will be set to 1.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Finally, note that while most feature concepts are lexicalized, others, such as the suffix concept, are not.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Because we don\u2019t have a separate development set, we used the training set to select among them and found 0.2 to work slightly better than 0.1 and 0.3.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Figure 4 shows the seven general knowledge sources (KSs) that represent features commonly used for coreference resolution.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Table 2 compares the performance of our system on the setup of Cohen and Smith (2007) to the best results reported by them for the same tasks.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Historically, Arabic grammar has identified two sentences types: those that begin with a nominal (\u00ef\u00bf\u00bd '.i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Assume that the two classifiers are &quot;rote learners&quot;: that is, 1.1 and 12 are defined through look-up tables that list a label for each member of X1 or X2.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"However, such an approach requires a number of hand-coded systems, which may not be available in languages other than English.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":", Sun day, then the feature DayOfTheWeek is set to 1.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Instead of the names of elementary trees of a TAG, the nodes are labeled by a sequence of names of trees in an elementary tree set.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"com t 600 Mountain Avenue, 2c278, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"However, there are phrases which express the same meanings even though they do not share the same keyword.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"To differentiate between the coordinating and discourse separator functions of conjunctions (Table 3), we mark each CC with the label of its right sister (splitCC).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Previous work deals with this problem by correcting inconsistencies between the named entity classes assigned to different occurrences of the same entity (Borthwick, 1999; Mikheev et al., 1998).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"(2009).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The Berkeley parser gives state-of-the-art performance for all metrics.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Trees are composed using an operation called adjoining, which is defined as follows.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The role that each noun phrase plays in the kidnapping event is key to distinguishing these cases.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"This result suggests the benefit of using the automatic discovery method.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Within this framework, we use features intended to capture degree of generality, including the output from an SVM classifier that uses the intersection between IN and OUT as positive examples.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Also, \u00e2\u0080\u009cagree\u00e2\u0080\u009d in the CC-domain is not a desirable keyword.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The relevant variables are the set of token-level tags that appear before and after each instance of the ith word type; we denote these context pairs with the set {(tb, ta)} and they are contained in t(\u00e2\u0088\u0092i).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As the reviewer also points out, this is a problem that is shared by, e.g., probabilistic context-free parsers, which tend to pick trees with fewer nodes.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"In this way we restrict the parameterization of a Language Original case English Danish Dutch German Spanish Swedish Portuguese 94.6 96.3 96.6 95.5 95.4 93.3 95.6 Table 1: Upper bound on tagging accuracy assuming each word type is assigned to majority POS tag.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Each learner is free to pick the labels for these instances.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"T e r r o r i s m Ca sef ra me Semantic Classes <a ge nt > ass ass ina ted group, human inv esti gat ion int o < N P> event exp lod ed out sid e < N P> building N a t u r a l D i s a s t e r s Ca sef ra me Semantic Classes <a ge nt > inv esti gat ing cau se group, human sur viv or of < N P> event, natphenom hit wit h < N P> attribute, natphenom Figure 3: Semantic Caseframe Expectations Figure 2: Lexical Caseframe Expectations To illustrate how lexical expectations are used, suppose we want to determine whether noun phrase X is the antecedent for noun phrase Y. If they are coreferent, then X and Y should be substitutable for one another in the story.4 Consider these sentences: (S1) Fred was killed by a masked man with a revolver.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"60 ( P e tr o v, 2 0 0 9 ) all B e r k e l e y ( S e p . 0 9 ) B a s e l i n e 7 0 a l l G o l d P O S 70 \u00e2\u0080\u0094 \u00e2\u0080\u0094 \u00e2\u0080\u0094 0 . 8 0 9 0.839 335 0 . 7 9","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"There are two weaknesses in Chang et al.'s model, which we improve upon.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Although the best published results for the Collins parser is 80% UAS (Collins, 1999), this parser reaches 82% when trained on the entire training data set, and an adapted version of Charniak\u2019s parser (Charniak, 2000) performs at 84% (Jan Haji\u02c7c, pers. comm.).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Finally, we note that Jiang\u2019s instance-weighting framework is broader than we have presented above, encompassing among other possibilities the use of unlabelled IN data, which is applicable to SMT settings where source-only IN corpora are available.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We asked six native speakers-three from Taiwan (TlT3), and three from the Mainland (M1M3)-to segment the corpus.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"There are still some open issues to be resolved with the format, but it represents a first step.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The belief value that would have been assigned to the intersection of these sets is .60*.70=.42, but this belief has nowhere to go because the null set is not permissible in the model.7 So this probability mass (.42) has to be redistributed.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The PROBING data structure is a rather straightforward application of these hash tables to store Ngram language models.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In this work we extended the AdaBoost.MH (Schapire and Singer 98) algorithm to the cotraining case.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The type-level tag assignments T generate features associated with word types W . The tag assignments constrain the HMM emission parameters \u00ce\u00b8.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The probabilities are incorporated into the DempsterShafer model using Equation 1.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Surprisingly, the non-parametric switching technique also exhibited robust behaviour in this situation.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Nonetheless, parse quality is much lower in the joint model because a lattice is effectively a long sentence.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"gaolbu4-gaolxing4 (hap-not-happy) 'happy?'","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Evaluation of Morphological Analysis.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"As expected, the vanilla HMM trained with EM performs the worst.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Considerable amount of work has been done in recent years on the named entity recognition task, partly due to the Message Understanding Conferences (MUC).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Most languages that use Roman, Greek, Cyrillic, Armenian, or Semitic scripts, and many that use Indian-derived scripts, mark orthographic word boundaries; however, languages written in a Chinese-derived writ\u00c2\u00ad ing system, including Chinese and Japanese, as well as Indian-derived writing systems of languages like Thai, do not delimit orthographic words.1 Put another way, written Chinese simply lacks orthographic words.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(Blum and Mitchell 98) give an example that illustrates just how powerful the second constraint can be.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"There has also been some work on adapting the word alignment model prior to phrase extraction (Civera and Juan, 2007; Wu et al., 2005), and on dynamically choosing a dev set (Xu et al., 2007).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"These are written to the state s(wn1) and returned so that they can be used for the following query.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In practice, this sparsity constraint is difficult to incorporate in a traditional POS induction system (Me\u00c2\u00b4rialdo, 1994; Johnson, 2007; Gao and Johnson, 2008; Grac\u00c2\u00b8a et al., 2009; Berg-Kirkpatrick et al., 2010).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(e.g., N.Y. would contribute this feature, IBM would not). nonalpha=x Appears if the spelling contains any characters other than upper or lower case letters.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Linear probing hash tables must have more buckets than entries, or else an empty bucket will never be found.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"6One of our experimental settings lacks document boundaries, and we used this approximation in both settings for consistency.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Evalb is a Java re-implementation of the standard labeled precision\/recall metric.12 The ATB gives all punctuation a single tag.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Using structural information As was explained in the results section, we extracted examples like \u00e2\u0080\u009cSmith estimates Lotus\u00e2\u0080\u009d, from a sentence like \u00e2\u0080\u009cMr.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"If one is interested in translation, one would probably want to consider show up as a single dictionary word since its semantic interpretation is not trivially derivable from the meanings of show and up.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"This paper is based on work supported in part by DARPA through IBM.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(1991}, Gu and Mao (1994), and Nie, Jin, and Hannan (1994).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Precision.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Although the segmentation requirements for Arabic are not as extreme as those for Chinese, Arabic is written with certain cliticized prepositions, pronouns, and connectives connected to adjacent words.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This implies, therefore, that a major factor in the performance of a Chinese segmenter is the quality of the base dictionary, and this is probably a more important factor-from the point of view of performance alone-than the particular computational methods used.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The method being described-henceforth ST..","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Note that hanzi that are not grouped into dictionary words (and are not identified as single\u00c2\u00ad hanzi words), or into one of the other categories of words discussed in this paper, are left unattached and tagged as unknown words.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"As one can see, only the trigrams [suo incarceramento ,], [suo iter ,] and [suo carattere ,] are connected to English words.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The maximum precision oracle is an upper bound on the possible gain we can achieve by parse hybridization.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"5 We choose these two metrics over the Variation Information measure due to the deficiencies discussed in Gao and Johnson (2008).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"JA DE G O L D G R AS S SI C K NE SS DE AT H R A T 14.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For the purposes of EM, the &quot;observed&quot; data is {(xi, Ya\u2022 \u2022 \u2022 (xrn, Yrn), xfil, and the hidden data is {ym+i y}.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"63 95.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"It is formally straightforward to extend the grammar to include these names, though it does increase the likelihood of overgeneration and we are unaware of any working systems that incorporate this type of name.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Such constraints are derived from training data, expressing some relationship between features and outcome.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"our full model yields 39.3% average error reduction across languages when compared to the basic configuration (1TW).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Different annotations of the same text are mapped into the same data structure, so that search queries can be formulated across annotation levels.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"However, for multinomial models like our LMs and TMs, there is a one to one correspondence between instances and features, eg the correspondence between a phrase pair (s, t) and its conditional multinomial probability p(s1t).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"JI!","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Discriminative Instance Weighting for Domain Adaptation in Statistical Machine Translation","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Our coreference resolver performed well in two domains, and experiments showed that each contextual role knowledge source contributed valuable information.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Our parsing performance measures (SY N) thus report the PARSEVAL extension proposed in Tsarfaty (2006).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"This approach makes the training objective more complex by adding linear constraints proportional to the number of word types, which is rather prohibitive.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"NN \u00ef\u00bf\u00bd .e NP NNP NP DTNNP NN \u00ef\u00bf\u00bd .e NP NP NNP NP Table 5: Evaluation of 100 randomly sampled variation nuclei types.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Mutual information was shown to be useful in the segmentation task given that one does not have a dictionary.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The state in future has not enough work for its many teachers.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We carried out translation experiments in two different settings.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"To our knowledge, ours is the first analysis of this kind for Arabic parsing.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"92 77.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Our clue is the NE instance pairs.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"To define a similarity function between the English and the foreign vertices, we rely on high-confidence word alignments.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"There has been recent interest in the application of Indexed Grammars (IG's) to natural languages.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Another approach to finding paraphrases is to find phrases which take similar subjects and objects in large corpora by using mutual information of word distribution [Lin and Pantel 01].","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We make the assumption that for each example, both xi,. and x2,2 alone are sufficient to determine the label yi.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"86 78.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Comparison with state-of-the-art taggers For comparison we consider two unsupervised tag- gers: the HMM with log-linear features of Berg- Kirkpatrick et al.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Final (F): The rest of the sentence is processed monotonically taking account of the already covered positions.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Apart from MERT difficulties, a conceptual problem with log-linear combination is that it multiplies feature probabilities, essentially forcing different features to agree on high-scoring candidates.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"7.96 5.55 1 l...................................................................................................................................................................................................J..","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"BerkeleyLM revision 152 (Pauls and Klein, 2011) implements tries based on hash tables and sorted arrays in Java with lossy quantization.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"In this paper, we make a simplifying assumption of one-tag-per-word.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The 95% confidence intervals for type-level errors are (5580, 9440) for the ATB and (1400, 4610) for the WSJ.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"To prevent this we &quot;smooth&quot; the confidence by adding a small value, e, to both W+ and W_, giving at = Plugging the value of at from Equ.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We attain these results using several optimizations: hashing, custom lookup tables, bit-level packing, and state for left-to-right query patterns.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"5 Related Work.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Their work used subject-verb, verb-object, and adjective-noun relations to compare the contexts surrounding an anaphor and candidate.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"raphy: A ren2 'person' is a fairly uncontroversial case of a monographemic word, and rplil zhong1guo2 (middle country) 'China' a fairly uncontroversial case of a di\u00c2\u00ad graphernic word.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As the reviewer also points out, this is a problem that is shared by, e.g., probabilistic context-free parsers, which tend to pick trees with fewer nodes.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"2.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We use v1.0 mainly because previous studies on joint inference reported results w.r.t. v1.0 only.5 We expect that using the same setup on v2.0 will allow a crosstreebank comparison.6 We used the first 500 sentences as our dev set and the rest 4500 for training and report our main results on this split.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Chris Dyer integrated the code into cdec.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kind of cue.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For instance, for TTS it is necessary to know that a particular sequence of hanzi is of a particular category because that knowl\u00c2\u00ad edge could affect the pronunciation; consider, for example the issues surrounding the pronunciation of ganl I qian2 discussed in Section 1.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"(2009).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For RandLM, we used the settings in the documentation: 8 bits per value and false positive probability 1 256.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"It was filtered to retain the top 30 translations for each source phrase using the TM part of the current log-linear model.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We quantify error categories in both evaluation settings.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The learned information was recycled back into the resolver to improve its performance.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The resulting algorithm is depicted in Table 1.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Yet, some hanzi are far more probable in women's names than they are in men's names, and there is a similar list of male-oriented hanzi: mixing hanzi from these two lists is generally less likely than would be predicted by the independence model.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The algorithm in Fig.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Models To assess the marginal utility of each component of the model (see Section 3), we incremen- tally increase its sophistication.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"An easy way to achieve this is to put the domain-specific LMs and TMs into the top-level log-linear model and learn optimal weights with MERT (Och, 2003).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The key to the methods we describe is redundancy in the unlabeled data.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The binary language model from Section 5.2 and text phrase table were forced into disk cache before each run.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The first issue relates to the completeness of the base lexicon.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"RandLM 0.2 (Talbot and Osborne, 2007) stores large-scale models in less memory using randomized data structures.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Link phrases based on instance pairs Using NE instance pairs as a clue, we find links between sets of phrases.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Typical data structures are generalized Bloom filters that guarantee a customizable probability of returning the correct answer.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Using the virtual distribution Di (i) and pseudo-labels&quot;y.,\u201e values for Wo, W\u00b1 and W_ can be calculated for each possible weak hypothesis (i.e., for each feature x E Xi); the weak hypothesis with minimal value for Wo + 2\/WW _ can be chosen as before; and the weight for this weak hypothesis at = ln ww+411:) can be calculated.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"A secondary reference resolution classifier has information on the class assigned by the primary classifier.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Several coreference resolvers have used supervised learning techniques, such as decision trees and rule learners (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, BABAR learned that agents that \u00e2\u0080\u009cassassinate\u00e2\u0080\u009d or \u00e2\u0080\u009cinvestigate a cause\u00e2\u0080\u009d are usually humans or groups (i.e., organizations).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"02 99.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Since Dane\u00cb\u0087s\u00e2\u0080\u0099 proposals of \u00e2\u0080\u0098thematic development patterns\u00e2\u0080\u0099, a few suggestions have been made as to the existence of a level of discourse structure that would predict the information structure of sentences within texts.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We use the default inference parameters.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Nonstochastic lexical-knowledge-based approaches have been much more numer\u00c2\u00ad ous.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If it is made up of all capital letters, then (allCaps, zone) is set to 1.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Evaluation of links A link between two sets is considered correct if the majority of phrases in both sets have the same meaning, i.e. if the link indicates paraphrase.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The 13 positions of the source sentence are processed in the order shown.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"On the other hand, if it is seen as McCann Pte.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"As Figure 1 shows, this word has no high-confidence alignment in the Italian-English bitext.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Mikheev et al.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Figure 1 depicts the lattice for a 2-words sentence bclm hneim.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In this paper k = 3 (the three labels are person, organization, location), and we set a = 0.1.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We propose an unsupervised method to discover paraphrases from a large untagged corpus, without requiring any seed phrase or other cue.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Many morphological decisions are based on long distance dependencies, and when the global syntactic evidence disagrees with evidence based on local linear context, the two models compete with one another, despite the fact that the PCFG takes also local context into account.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Turning now to (1), we have the similar problem that splitting.into.ma3 'horse' andlu4 'way' is more costly than retaining this as one word .ma3lu4 'road.'","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The trie data structure is commonly used for language modeling.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Finally, we show that in application settings, the absence of gold segmentation lowers parsing performance by 2\u00e2\u0080\u00935% F1.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"First, we aim to explicitly characterize examples from OUT as belonging to general language or not.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"3.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Global context from the whole document is available and can be exploited in a natural manner with a maximum entropy classifier.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"A dynamic programming recursion similar to the one in Eq. 2 is evaluated.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In our experiments we set the parameter values randomly, and then ran EM to convergence.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Then we ran binary search to determine the least amount of memory with which it would run.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We use the log-linear tagger of Toutanova et al.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"As shown in Table 3, the proportion of sentences containing some non-projective dependency ranges from about 15% in DDT to almost 25% in PDT.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Each xt E 2x is the set of features constituting the ith example.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Schapire and Singer show that the training error is bounded above by Thus, in order to greedily minimize an upper bound on training error, on each iteration we should search for the weak hypothesis ht and the weight at that minimize Z.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Unigram records store probability, backoff, and an index in the bigram table.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"To eliminate such sequences, we define a transition probability between word classes to be equal to 1 if the sequence is admissible, and 0 otherwise.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This supports our main thesis that decisions taken by single, improved, grammar are beneficial for both tasks.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We identified three ways that contextual roles can be exploited: (1) by identifying caseframes that co-occur in resolutions, (2) by identifying nouns that co-occur with case- frames and using them to crosscheck anaphor\/candidate compatibility, (3) by identifying semantic classes that co- occur with caseframes and using them to crosscheck anaphor\/candidate compatability.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"A modified language model probability p\u00c3\u0086(eje0; e00) is defined as follows: p\u00c3\u0086(eje0; e00) =  1:0 if \u00c3\u0086 = 0 p(eje0; e00) if \u00c3\u0086 = 1 : We associate a distribution p(\u00c3\u0086) with the two cases \u00c3\u0086 = 0 and \u00c3\u0086 = 1 and set p(\u00c3\u0086 = 1) = 0:7.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Hence, we take the probability of the event fmnh analyzed as REL VB to be This means that we generate f and mnh independently depending on their corresponding PoS tags, and the context (as well as the syntactic relation between the two) is modeled via the derivation resulting in a sequence REL VB spanning the form fmnh. based on linear context.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Entries for 2 < n < N store a vocabulary identifier, probability, backoff, and an index into the n + 1-gram table.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This feature imposes constraints Table 1: Features based on the token string that are based on the probability of each name class during training.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"For example, we might have VP \u00e2\u0086\u0092 VB NP PP, where the NP is the subject.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In our coreference resolver, we define \u00ce\u00b8 to be the set of all candidate antecedents for an anaphor.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The less favored reading may be selected in certain contexts, however; in the case of , for example, the nominal reading jiang4 will be selected if there is morphological information, such as a following plural affix ir, menD that renders the nominal reading likely, as we shall see in Section 4.3.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"7).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We tokenize MWUs and their POS tags; this reduces the tag set size to 12.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The two new terms force the two classifiers to agree, as much as possible, on the unlabeled examples.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Specifically, for the ith word type, the set of token-level tags associated with token occurrences of this word, denoted t(i), must all take the value Ti to have nonzero mass. Thus in the context of Gibbs sampling, if we want to block sample Ti with t(i), we only need sample values for Ti and consider this setting of t(i).","label":0},{"source_text":"Here we present two algorithms.","target_text":"Taking only the highest frequency rules is much &quot;safer&quot;, as they tend to be very accurate.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In the case of adverbial reduplication illustrated in (3b) an adjective of the form AB is reduplicated as AABB.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We extend the Stanford parser to accept pre-generated lattices, where each word is represented as a finite state automaton.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(3) shows learning curves for CoBoost.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"All features were conjoined with the state z.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Le m\u00b4edicament de r\u00b4ef\u00b4erence de Silapo est EPREX\/ERYPO, qui contient de l\u2019\u00b4epo\u00b4etine alfa.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"During training, we treat as observed the language word types W as well as the token-level corpus w. We utilize Gibbs sampling to approximate our collapsed model posterior: P (T ,t|W , w, \u00ce\u00b1, \u00ce\u00b2) \u00e2\u0088\u009d P (T , t, W , w|\u00ce\u00b1, \u00ce\u00b2) 0.7 0.6 0.5 0.4 0.3 English Danish Dutch Germany Portuguese Spanish Swedish = P (T , t, W , w, \u00cf\u0088, \u00ce\u00b8, \u00cf\u0086, w|\u00ce\u00b1, \u00ce\u00b2)d\u00cf\u0088d\u00ce\u00b8d\u00cf\u0086 Note that given tag assignments T , there is only one setting of token-level tags t which has mass in the above posterior.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Their theoretical finding is simply stated: classification error rate decreases toward the noise rate exponentially in the number of independent, accurate classifiers.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The 8 similarity-to-IN features are based on word frequencies and scores from various models trained on the IN corpus: To avoid numerical problems, each feature was normalized by subtracting its mean and dividing by its standard deviation.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"This annotation choice weakens splitIN.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Besides the lack of a clear definition of what constitutes a correct segmentation for a given Chinese sentence, there is the more general issue that the test corpora used in these evaluations differ from system to system, so meaningful comparison between systems is rendered even more difficult.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Maamouri et al.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Ex: Mr. Cristiani, president of the country ...","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We see from these results that the behavior of the parametric techniques are robust in the presence of a poor parser.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"When the connective is an adverbial, there is much less clarity as to the range of the spans.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"P St = n. \u00ce\u00b2 T VARIABLES \u00cf\u0088 Y W : Word types (W1 ,.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"However, it can be noted that the results for the least informative encoding, Path, are almost comparable, while the third encoding, Head, gives substantially worse results for both data sets.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This representation gives ir, an appropriate morphological decomposition, pre\u00c2\u00ad serving information that would be lost by simply listing ir, as an unanalyzed form.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We tokenize MWUs and their POS tags; this reduces the tag set size to 12.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"As mentioned above, it is not obvious how to apply Daum\u00b4e\u2019s approach to multinomials, which do not have a mechanism for combining split features.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Semantic expectations are analogous to lexical expectations except that they represent semantic classes rather than nouns.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This method, one instance of which we term the \"greedy algorithm\" in our evaluation of our own system in Section 5, involves starting at the beginning (or end) of the sentence, finding the longest word starting (ending) at that point, and then repeating the process starting at the next (previous) hanzi until the end (begin\u00c2\u00ad ning) of the sentence is reached.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"State will ultimately be used as context in a subsequent query.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"In such cases we use the non-pruned lattice including all (possibly ungrammatical) segmentation, and let the statistics (including OOV) decide.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For t = 1, T and for j = 1, 2: where 4 = exp(-jg'(xj,i)). practice, this greedy approach almost always results in an overall decrease in the value of Zco.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Figure 4 shows some such phrase sets based on keywords in the CC-domain.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"9 66.4 47.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"But we will show that the use of unlabeled data can drastically reduce the need for supervision.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The second modification is more important, and is discussed in the next section.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Hereafter, each pair of NE categories will be called a domain; e.g. the \u00e2\u0080\u009cCompany \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain, which we will call CC- domain (Step 2).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Bresnan, Kaplan, Peters, and Zaenen (1982) argue that these structures are needed to describe crossed-serial dependencies in Dutch subordinate clauses.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"If these sets do not overlap, then the words cannot be coreferent.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Thus, the effects of spontaneous speech are present in the corpus, e.g. the syntactic structure of the sentence is rather less restricted, however the effect of speech recognition errors is not covered.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The authors acknowledge the support of the NSF (CAREER grant IIS0448168, and grant IIS 0904684).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"We use MMAX for this annotation as well.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"paper, and is missing 6 examples from the A set.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The final model tions.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The second case involves existential noun phrases (Allen, 1995), which are noun phrases that uniquely specify an object or concept and therefore do not need a prior referent in the discourse.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In sequential tagging models such as (Adler and Elhadad, 2006; Bar-Haim et al., 2007; Smith et al., 2005) weights are assigned according to a language model The input for the joint task is a sequence W = w1, ... , wn of space-delimited tokens.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Limitations of (Blum and Mitchell 98): While the assumptions of (Blum and Mitchell 98) are useful in developing both theoretical results and an intuition for the problem, the assumptions are quite limited.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Also, the argument has been made that machine translation performance should be evaluated via task-based evaluation metrics, i.e. how much it assists performing a useful task, such as supporting human translators or aiding the analysis of texts.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In Eq.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Let us notate the set of previously unseen, or novel, members of a category X as unseen(X); thus, novel members of the set of words derived in f, menO will be de\u00c2\u00ad noted unseen(f,).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This suggests that different types of anaphora may warrant different treatment: definite NP resolution may depend more on lexical semantics, while pronoun resolution may depend more on contextual semantics.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"2.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"As noted for the perplexity task, we do not expect cache to grow substantially with model size, so RandLM remains a low-memory option.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Note that in line 4 the last visited position for the successor hypothesis must be m. Otherwise , there will be four uncovered positions for the predecessor hypothesis violating the restriction.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Third, we develop a human interpretable grammar that is competitive with a latent variable PCFG.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"A number of PCC commentaries will be read by professional news speakers and prosodic features be annotated, so that the various annotation layers can be set into correspondence with intonation patterns.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"This class of formalisms have the properties that their derivation trees are local sets, and manipulate objects, using a finite number of composition operations that use a finite number of symbols.","label":0},{"source_text":"There is no global pruning.","target_text":"For each extension a new position is added to the coverage set.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(1992).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"9 66.4 47.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"La ng ua ge 1T W + P RI O R + F E A T S E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 2 1.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"For example, Gazdar (1985) discusses the applicability of Indexed Grammars (IG's) to Natural Language in terms of the structural descriptions assigned; and Berwick (1984) discusses the strong generative capacity of Lexical-Functional Grammar (LFG) and Government and Bindings grammars (GB).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For the examples given in (1) and (2) this certainly seems possible.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We applied the AutoSlog system (Riloff, 1996) to our unannotated training texts to generate a set of extraction patterns for each domain.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Overview of the method 2.2 Step by Step Algorithm.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"12 For English, our Evalb implementation is identical to the most recent reference (EVALB20080701).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"3 Throughout this paper we shall give Chinese examples in traditional orthography, followed.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(a) IDictionary D I D:d\/0.000 B:b\/0.000 B:b\/0.000 ( b ) ( c ) ( d ) I B e s t P a t h ( I d ( I ) o D * ) I cps:nd4.!l(l() Figure 2 An abstract example illustrating the segmentation algorithm.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Out-of-Vocabulary: We derived a lexicon list from WordNet 1.6, and words that are not found in this list have a feature out-of-vocabulary set to 1.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"By contrast, BerkeleyLM\u2019s hash and compressed variants will return incorrect results based on an n \u22121-gram.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Only one of the first n positions which are not already aligned in a partial hypothesis may be chosen, where n is set to 4.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"95 B a s e li n e ( S e lf t a g ) 70 a l l B i k e l ( v 1 . 2 ) B a s e l i n e ( P r e t a g ) 7 0 a l l G o l d P O S 70 0.7 70 0.801 278 0.7 52 0.794 278 0.7 71 0.804 295 0.7 52 0.796 295 0.7 75 0.808 309 77.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We showed that any system defined in this way can be recognized in polynomial time.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Unexpectedly, because the ministries of treasury and education both had prepared the teacher plan together.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The errors shown are from the Berkeley parser output, but they are representative of the other two parsing models.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Therefore, the number of fine tags varied across languages for our experiments; however, one could as well have fixed the set of HMM states to be a constant across languages, and created one mapping to the universal POS tagset.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Its correct antecedent is \u00e2\u0080\u009ca revolver\u00e2\u0080\u009d, which is extracted by the caseframe \u00e2\u0080\u009ckilled with <NP>\u00e2\u0080\u009d.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Previous work deals with this problem by correcting inconsistencies between the named entity classes assigned to different occurrences of the same entity (Borthwick, 1999; Mikheev et al., 1998).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, if CFLex determines that the log- likelihood statistic for the co-occurrence of a particular noun and caseframe corresponds to the 90% confidence level, then CFLex returns .90 as its belief that the anaphor and candidate are coreferent.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In any event, to date, we have not compared different methods for deriving the set of initial frequency estimates.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Examples are given in Table 4.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Thus corresponding pseudo-labels for instances on which gj abstain are set to zero and these instances do not contribute to the objective function.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"4.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We simulate lexical constraints by using an external lexical resource against which we verify whether OOV segments are in fact valid Hebrew lexemes.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The alignment mapping is j ! i = aj from source position j to target position i = aj . The use of this alignment model raises major problems if a source word has to be aligned to several target words, e.g. when translating German compound nouns.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"(2) Description Source Location Names http:\/\/www.timeanddate.com http:\/\/www.cityguide.travel-guides.com http:\/\/www.worldtravelguide.net Corporate Names http:\/\/www.fmlx.com Person First Names http:\/\/www.census.gov\/genealogy\/names Person Last Names Table 2: Sources of Dictionaries The McCann family . . .","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For a description of the application of AdaBoost to various NLP problems see the paper by Abney, Schapire, and Singer in this volume.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The second modification is more important, and is discussed in the next section.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"However, such an approach requires a number of hand-coded systems, which may not be available in languages other than English.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Both authors are members of the Center for Language and Speech Processing at Johns Hopkins University.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We develop our POS induction model based on the feature-based HMM of Berg-Kirkpatrick et al. (2010).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In Section 3, we introduce our novel concept to word reordering and a DP-based search, which is especially suitable for the translation direction from German to English.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The features are weighted within a logistic model to give an overall weight that is applied to the phrase pair\u2019s frequency prior to making MAP-smoothed relative-frequency estimates (different weights are learned for each conditioning direction).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Table 8b shows that verbal nouns are the hardest pre-terminal categories to identify.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Annotation of syntactic structure for the core corpus has just begun.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"RandLM\u2019s stupid backoff variant stores counts instead of probabilities and backoffs.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Tokens were converted to vocabulary identifiers in advance and state was carried from each query to the next.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The basic strategy is, for a given pair of entity types, to start with some examples, like several famous book title and author pairs; and find expressions which contains those names; then using the found expressions, find more author and book title pairs.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The major problem for our seg\u00c2\u00ad menter, as for all segmenters, remains the problem of unknown words (see Fung and Wu [1994]).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"This is not an oversight.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In order to solve this problem, a parse tree is needed to understand that \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is not the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, until such standards are universally adopted in evaluating Chinese segmenters, claims about performance in terms of simple measures like percent correct should be taken with a grain of salt; see, again, Wu and Fung (1994) for further arguments supporting this conclusion.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The method reported in this paper makes use solely of unigram probabilities, and is therefore a zeroeth-order model: the cost of a particular segmentation is estimated as the sum of the costs of the individual words in the segmentation.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The second setting uses the news-related subcorpora for the NIST09 MT Chinese to English evaluation8 as IN, and the remaining NIST parallel Chinese\/English corpora (UN, Hong Kong Laws, and Hong Kong Hansard) as OUT.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Input: Ja , wunderbar . Konnen wir machen . MonS: Yes, wonderful.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In addition we report for each model its performance on goldsegmented input (GS) to indicate the upper bound 11Overt definiteness errors may be seen as a wrong feature rather than as wrong constituent and it is by now an accepted standard to report accuracy with and without such errors. for the grammars\u2019 performance on the parsing task.","label":0},{"source_text":"There is no global pruning.","target_text":"In.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"(1), Pr(eI 1) is the language model, which is a trigram language model in this case.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Table 8a shows that the best model recovers SBAR at only 71.0% F1.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We allow any number of bits from 2 to 25, unlike IRSTLM (8 bits) and BerkeleyLM (17\u221220 bits).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Second, we treat the projected labels as features in an unsupervised model (\u00a75), rather than using them directly for supervised training.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Our method does not assume any knowledge about the target language (in particular no tagging dictionary is assumed), making it applicable to a wide array of resource-poor languages.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The alignment mapping is j ! i = aj from source position j to target position i = aj . The use of this alignment model raises major problems if a source word has to be aligned to several target words, e.g. when translating German compound nouns.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The input to AdaBoost is a set of training examples ((xi , yi), , (x\u201e.\u201e yrn)).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"2.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, kidnapping victims should be extracted from the subject of the verb \u00e2\u0080\u009ckidnapped\u00e2\u0080\u009d when it occurs in the passive voice (the shorthand representation of this pattern would be \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We will show that by giving the first model some global features, MENERGI outperforms Borthwick' s reference resolution classifier.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"If the expression is a word or a short phrase (like \u00e2\u0080\u009ccorporation\u00e2\u0080\u009d and \u00e2\u0080\u009ccompany\u00e2\u0080\u009d), it is called a \u00e2\u0080\u009csynonym\u00e2\u0080\u009d.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The unlabeled data gives many such &quot;hints&quot; that two features should predict the same label, and these hints turn out to be surprisingly useful when building a classifier.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This intuition is born out by the experimental results.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Secondly, for all our models we provide better fine- and coarse-grained POS-tagging accuracy, and all pruned models outperform the Oracle results reported by them.12 In terms of syntactic disambiguation, even the simplest grammar pruned with HSPELL outperforms their non-Oracle results.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"La ng ua ge # To ke ns # W or d Ty pe s # Ta gs E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 1 1 7 3 7 6 6 9 4 3 8 6 2 0 3 5 6 8 6 9 9 6 0 5 2 0 6 6 7 8 8 9 3 3 4 1 9 1 4 6 7 4 9 2 0 6 1 8 3 5 6 2 8 3 9 3 7 2 3 2 5 2 8 9 3 1 1 6 4 5 8 2 0 0 5 7 4 5 2 5 1 2 5 4 2 2 4 7 4 1 Table 2: Statistics for various corpora utilized in experiments.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Each unlabeled pair (x1,i, x2,i) is represented as an edge between nodes corresponding to x1,i and X2,i in the graph.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"As indicated by bolding, for seven out of eight languages the improvements of the \u201cWith LP\u201d setting are statistically significant with respect to the other models, including the \u201cNo LP\u201d setting.11 Overall, it performs 10.4% better than the hitherto state-of-the-art feature-HMM baseline, and 4.6% better than direct projection, when we macro-average the accuracy over all languages.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"This is manifest in the lexical choices but 1 www.coli.unisb.de\/\u00e2\u0088\u00bcthorsten\/tnt\/ Dagmar Ziegler is up to her neck in debt.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Adam Pauls provided a pre-release comparison to BerkeleyLM and an initial Java interface.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We further thank Dr. J.-S.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The procedure using the tagged sentences to discover paraphrases takes about one hour on a 2GHz Pentium 4 PC with 1GB of memory.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Since pronouns carry little semantics of their own, resolving them depends almost entirely on context.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This WFST represents the segmentation of the text into the words AB and CD, word boundaries being marked by arcs mapping between f and part-of-speech labels.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Second, we show that although the Penn Arabic Treebank is similar to other tree- banks in gross statistical terms, annotation consistency remains problematic.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The dev and test sets were randomly chosen from the EMEA corpus.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"SRILM\u2019s compact variant has an incredibly expensive destructor, dwarfing the time it takes to perform translation, and so we also modified Moses to avoiding the destructor by calling exit instead of returning normally.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Part-of-speech (POS) tag distributions are known to exhibit sparsity \u00e2\u0080\u0094 a word is likely to take a single predominant tag in a corpus.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In this section, we brie y review our translation approach.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"As shown in Table 3, the proportion of sentences containing some non-projective dependency ranges from about 15% in DDT to almost 25% in PDT.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We first make use of our morphological analyzer to find all segmentation possibilities by chopping off all prefix sequence possibilities (including the empty prefix) and construct a lattice off of them.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"(2009).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"(2010) reports the best unsupervised results for English.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Trying to integrate constituent ordering and choice of referring expressions, (Chiarcos 2003) developed a numerical model of salience propagation that captures various factors of author\u00e2\u0080\u0099s intentions and of information structure for ordering sentences as well as smaller constituents, and picking appropriate referring expressions.10 Chiarcos used the PCC annotations of co-reference and information structure to compute his numerical models for salience projection across the generated texts.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In contrast to the Bayesian HMM, \u00ce\u00b8t is not drawn from a distribution which has support for each of the n word types.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The system will then look for sequences of initial capitalized words that match the acronyms found in the whole document.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"We follow the guidelines developed in the TIGER project (Brants et al. 2002) for syntactic annotation of German newspaper text, using the Annotate3 tool for interactive construction of tree structures.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Another way to interpret this is that less than 5% of the correct constituents are missing from the hypotheses generated by the union of the three parsers.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Tables 4 and 5 also show that putting all of the contextual role KSs in play at the same time produces the greatest performance gain.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Covering the first uncovered position in the source sentence, we use the language model probability p(ej$; $).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The results of this experiment can be seen in Table 5.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Learned Tag Prior (PRIOR) We next assume there exists a single prior distribution \u00cf\u0088 over tag assignments drawn from DIRICHLET(\u00ce\u00b2, K ).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For definite NPs, the results are a mixed bag: some knowledge sources increased recall a little, but at the expense of some precision.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The availability of these resources guided our selection of foreign languages.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The second row represents the performance of the median hyperparameter setting.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"For developing these mechanisms, the possibility to feed in hand-annotated information is very useful.","label":0},{"source_text":"There is no global pruning.","target_text":"This approach leads to a search procedure with complexity O(E3 J4).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Now we have sets of phrases which share a keyword and we have links between those sets.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We combined evidence from four contextual role knowledge sources with evidence from seven general knowledge sources using a DempsterShafer probabilistic model.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Denote the unthresholded classifiers after t \u2014 1 rounds by git\u20141 and assume that it is the turn for the first classifier to be updated while the second one is kept fixed.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Thus an explicit assumption about the redundancy of the features \u2014 that either the spelling or context alone should be sufficient to build a classifier \u2014 has been built into the algorithm.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"None of the models we have presented utilize features associated with a particular constituent (i.e. the label, span, parent label, etc.) to influence parser preference.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Only one of the first n positions which are not already aligned in a partial hypothesis may be chosen, where n is set to 4.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"(Carlson, Marcu 2001) responded to this situation with relatively precise (and therefore long!)","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Queries take the form p(wn|wn\u22121 1 ) where wn1 is an n-gram.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In this section, we describe the seven general knowledge sources and explain how the DempsterShafer model makes resolutions.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The model described here thus demonstrates great potential for use in widespread applications.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Hence, there is a total of 29 classes (7 name classes 4 sub-classes 1 not-a-name class).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Data We use the Hebrew Treebank, (Sima\u2019an et al., 2001), provided by the knowledge center for processing Hebrew, in which sentences from the daily newspaper \u201cHa\u2019aretz\u201d are morphologically segmented and syntactically annotated.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"4.2 A Sample Segmentation Using Only Dictionary Words Figure 4 shows two possible paths from the lattice of possible analyses of the input sentence B X:\u00c2\u00a5 .:.S:P:l 'How do you say octopus in Japanese?' previously shown in Figure 1.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Yet we note that the better grammars without pruning outperform the poorer grammars using this technique, indicating that the syntactic context aids, to some extent, the disambiguation of unknown tokens.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"All commentaries have been tagged with part-of-speech information using Brants\u00e2\u0080\u0099 TnT1 tagger and the Stuttgart\/Tu\u00c2\u00a8bingen Tag Set automatic analysis was responsible for this decision.)","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Our experiments consistently demonstrate that this model architecture yields substantial performance gains over more complex tagging counterparts.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The compact variant uses sorted arrays instead of hash tables within each node, saving some memory, but still stores full 64-bit pointers.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Instead, we focused on phrases and set the frequency threshold to 2, and so were able to utilize a lot of phrases while minimizing noise.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For corporate suffixes, a list of tokens cslist that occur frequently as the last token of an organization name is collected from the training data.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For a given partial hypothesis (C; j), the order in which the cities in C have been visited can be ignored (except j), only the score for the best path reaching j has to be stored.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This representation gives ir, an appropriate morphological decomposition, pre\u00c2\u00ad serving information that would be lost by simply listing ir, as an unanalyzed form.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This is done using a simple PCFG which is lexemebased.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Parser 3, the most accurate parser, was chosen 71% of the time, and Parser 1, the least accurate parser was chosen 16% of the time.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Again, this deserves further investigation.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The hash variant is a reverse trie with hash tables, a more memory-efficient version of SRILM\u2019s default.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Given an anaphor, BABAR identifies the caseframe that would extract it from its sentence.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For all lists except locations, the lists are processed into a list of tokens (unigrams).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Figure 4 Input lattice (top) and two segmentations (bottom) of the sentence 'How do you say octopus in Japanese?'.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The number of latent HMM states for each language in our experiments was set to the number of fine tags in the language\u2019s treebank.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Extract NE pair instances with contexts From the four years of newspaper corpus, we extracted 1.9 million pairs of NE instances.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Recently, statistical NERs have achieved results that are comparable to hand-coded systems.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"An important reason for separating the two types of features is that this opens up the possibility of theoretical analysis of the use of unlabeled examples.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"In the rhetorical tree, nuclearity information is then used to extract a \u00e2\u0080\u009ckernel tree\u00e2\u0080\u009d that supposedly represents the key information from which the summary can be generated (which in turn may involve co-reference information, as we want to avoid dangling pronouns in a summary).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Acknowledgments We thank Steven Bethard, Evan Rosen, and Karen Shiells for material contributions to this work.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Brin 98) ,describes a system for extracting (author, book-title) pairs from the World Wide Web using an approach that bootstraps from an initial seed set of examples.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The feature-based model replaces the emission distribution with a log-linear model, such that: on the word identity x, features checking whether x contains digits or hyphens, whether the first letter of x is upper case, and suffix features up to length 3.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Moses sets the cache size parameter to 50 so we did as well; the resulting cache size is 2.82 GB.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"A direct-mapped cache makes BerkeleyLM faster on repeated queries, but their fastest (scrolling) cached version is still slower than uncached PROBING, even on cache-friendly queries.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"It was also proposed to allow annotators to skip sentences that they are unable to judge.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The number of latent HMM states for each language in our experiments was set to the number of fine tags in the language\u2019s treebank.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Ex: Mr. Cristiani is the president ...","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Finally, the concatenated 5 * 20% output is used to train the reference resolution component.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Instead, we resort to an iterative update based method.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"All improvements over the baseline are statistically significant beyond the 0.01 level (McNemar\u2019s test).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"These alignment models are similar to the concept of hidden Markov models (HMM) in speech recognition.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Rather we believe several methods have to be developed using different heuristics to discover wider variety of paraphrases.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The present proposal falls into the last group.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The developers aimed to reduce memory consumption at the expense of time.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Our second point of comparison is with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Whether a language even has orthographic words is largely dependent on the writing system used to represent the language (rather than the language itself); the notion \"orthographic word\" is not universal.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Such constraints are derived from training data, expressing some relationship between features and outcome.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The 2(Yarowsky 95) describes the use of more sophisticated smoothing methods.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"In focus is in particular the correlation with rhetorical structure, i.e., the question whether specific rhetorical relations \u00e2\u0080\u0094 or groups of relations in particular configurations \u00e2\u0080\u0094 are signalled by speakers with prosodic means.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Finally, we find links between sets of phrases, based on the NE instance pair data (for example, different phrases which link \u00e2\u0080\u009cIBM\u00e2\u0080\u009d and \u00e2\u0080\u009cLotus\u00e2\u0080\u009d) (Step 4).","label":0},{"source_text":"Here we present two algorithms.","target_text":"AdaBoost is given access to a weak learning algorithm, which accepts as input the training examples, along with a distribution over the instances.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Prague Dependency Treebank (Haji\u02c7c et al., 2001b), Danish Dependency Treebank (Kromann, 2003), and the METU Treebank of Turkish (Oflazer et al., 2003), which generally allow annotations with nonprojective dependency structures.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"So, there is a limitation that IE can only be performed for a predefined task, like \u00e2\u0080\u009ccorporate mergers\u00e2\u0080\u009d or \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Evalb, the standard parsing metric, is biased toward such corpora (Sampson and Babarczy, 2003).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"In addition, each feature function is a binary function.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"It should be clear from the onset that the particle b (\u201cin\u201d) in \u2018bcl\u2019 may then attach higher than the bare noun cl (\u201cshadow\u201d).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"A similar maximumlikelihood approach was used by Foster and Kuhn (2007), but for language models only.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In Eq.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"our full model yields 39.3% average error reduction across languages when compared to the basic configuration (1TW).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"971,746 sentences of New York Times text were parsed using the parser of (Collins 96).1 Word sequences that met the following criteria were then extracted as named entity examples: whose head is a singular noun (tagged NN).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We call this technique constituent voting.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We of course also fail to identify, by the methods just described, given names used without their associated family name.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"This work was funded by NSF grant IRI-9502312.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"These three parsers have given the best reported parsing results on the Penn Treebank Wall Street Journal corpus (Marcus et al., 1993).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The techniques we develop can be extended in a relatively straightforward manner to the more general case when OUT consists of multiple sub-domains.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Giving a recognition algorithm for LCFRL's involves describing the substrings of the input that are spanned by the structures derived by the LCFRS's and how the composition operation combines these substrings.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Therefore, the number of fine tags varied across languages for our experiments; however, one could as well have fixed the set of HMM states to be a constant across languages, and created one mapping to the universal POS tagset.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Our analysis and comparison focuses primarily on the one-to-one accuracy since it is a stricter metric than many-to-one accuracy, but also report many-to-one for completeness.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We now describe the CoBoost algorithm for the named entity problem.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Table 3 Classes of words found by ST for the test corpus.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"As seen by the drop in average individual parser performance baseline, the introduced parser does not perform very well.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Compared to related work on the recovery of long-distance dependencies in constituency-based parsing, our approach is similar to that of Dienes and Dubey (2003) in that the processing of non-local dependencies is partly integrated in the parsing process, via an extension of the set of syntactic categories, whereas most other approaches rely on postprocessing only.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"One may argue with these efforts on normalization, and ultimately their value should be assessed by assessing their impact on inter-annotator agreement.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The method uses a &quot;soft&quot; measure of the agreement between two classifiers as an objective function; we described an algorithm which directly optimizes this function.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In the following, we assume that this word joining has been carried out.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"SRILM\u2019s compact variant, IRSTLM, MITLM, and BerkeleyLM\u2019s sorted variant are all based on this technique.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Many statistical or machine-learning approaches for natural language problems require a relatively large amount of supervision, in the form of labeled training examples.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Lazy mapping reduces memory requirements by loading pages from disk only as necessary.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"This process produces a large set of caseframes coupled with a list of the noun phrases that they extracted.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"A corpus of German newspaper commentaries has been assembled and annotated with different information (and currently, to different degrees): part-of-speech, syntax, rhetorical structure, connectives, co-reference, and information structure.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For each language, we took the same number of sentences from the bitext as there are in its treebank, and trained a supervised feature-HMM.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Suffixes and Prefixes: This group contains only two features: Corporate-Suffix and Person-Prefix.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Next, for each pair of NE categories, we collect all the contexts and find the keywords which are topical for that NE category pair.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The original OUT counts co(s, t) are weighted by a logistic function w\u03bb(s, t): To motivate weighting joint OUT counts as in (6), we begin with the \u201cideal\u201d objective for setting multinomial phrase probabilities 0 = {p(s|t), dst}, which is the likelihood with respect to the true IN distribution pi(s, t).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"However there is no global pruning.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Figure 2 shows a screenshot (which is of somewhat limited value, though, as color plays a major role in signalling the different statuses of the information).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Such sequences are given additional features of A begin, A continue, or A end, and the acronym is given a feature A unique.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"For all grammars, we use fine-grained PoS tags indicating various morphological features annotated therein.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold, 30.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"When aligning the words in parallel texts (for language pairs like SpanishEnglish, French-English, ItalianGerman,...), we typically observe a strong localization effect.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"On a set of 11 sentence fragments-the A set-where they reported 100% recall and precision for name identification, we had 73% recall and 80% precision.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The first value reports resident size after loading; the second is the gap between post-loading resident memory and peak virtual memory.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"German city names are replaced by category markers.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The above auxiliary quantity satisfies the following recursive DP equation: Qe0 (e; C; j) = Initial Skip Verb Final 1.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Word Re-ordering and DP-based Search in Statistical Machine Translation","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Moreover, if we want to test on a huge test corpus, indexing the whole corpus might prove computationally expensive.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Like CFG's, the choice is predetermined by a finite number of rules encapsulated in the grammar.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Note that the sets of possible classifiers for a given noun can easily be encoded on that noun by grammatical features, which can be referred to by finite-state grammatical rules.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The composition operations are mapped onto operations that use concatenation to define the substrings spanned by the resulting structures.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Location list is processed into a list of unigrams and bigrams (e.g., New York).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This best instance-weighting model beats the equivalant model without instance weights by between 0.6 BLEU and 1.8 BLEU, and beats the log-linear baseline by a large margin.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"00 76.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The model described here thus demonstrates great potential for use in widespread applications.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Let s = a + b.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"3) A tight coupling with the speech recognizer output.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The linear and nonerasing assumptions about the operations discussed in Section 4.1 require that each z, and yk is used exactly once to define the strings zi, ,z1,3.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Personal names such as 00, 3R; zhoulenl-lai2 'Zhou Enlai.'","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"By using the output of a hand-coded system such as Proteus, MENE can improve its performance, and can even outperform IdentiFinder (Borthwick, 1999).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The system described in this paper is similar to the MENE system of (Borthwick, 1999).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"10 Here we use the Good-Turing estimate (Baayen 1989; Church and Gale 1991), whereby the aggregate probability of previously unseen instances of a construction is estimated as ni\/N, where N is the total number of observed tokens and n1 is the number of types observed only once.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Borth 2MUC data can be obtained from the Linguistic Data Consortium: http:\/\/www.ldc.upenn.edu 3Training data for IdentiFinder is actually given in words (i.e., 650K & 790K words), rather than tokens Table 6: Comparison of results for MUC7 wick (1999) reported using dictionaries of person first names, corporate names and suffixes, colleges and universities, dates and times, state abbreviations, and world regions.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"TIS systems in general need to do more than simply compute the.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For t = 1, T and for j = 1, 2: where 4 = exp(-jg'(xj,i)). practice, this greedy approach almost always results in an overall decrease in the value of Zco.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Without HSPELL-pruning, our simpler grammars are somewhat lagging behind, but as the grammars improve the gap is bridged.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Methods that allow multiple segmentations must provide criteria for choosing the best segmentation.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Our approach uses word-to-word dependencies between source and target words.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For each pair of judges, consider one judge as the standard,.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The tee in Figure 9a can be denoted by t1 i223 where tee substitution is used instead of concatenation.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Under certain conditions the constituent voting and na\u00efve Bayes constituent combination techniques are guaranteed to produce sets of constituents with no crossing brackets.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Linear weights are difficult to incorporate into the standard MERT procedure because they are \u201chidden\u201d within a top-level probability that represents the linear combination.1 Following previous work (Foster and Kuhn, 2007), we circumvent this problem by choosing weights to optimize corpus loglikelihood, which is roughly speaking the training criterion used by the LM and TM themselves.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"P St = n. \u00ce\u00b2 T VARIABLES \u00cf\u0088 Y W : Word types (W1 ,.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Next, we describe four contextual role knowledge sources that are created from the training examples and the caseframes.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Among these 32 sets, we found the following pairs of sets which have two or more links.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"A list of words occurring more than 10 times in the training data is also collected (commonWords).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In Semitic languages the situation is very different.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"4 65.9 48.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In MUC6, the best result is achieved by SRA (Krupka, 1995).","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In approaching this problem, a variety of different methods are conceivable, including a more or less sophisticated use of machine learning.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Compared with SRILM, IRSTLM adds several features: lower memory consumption, a binary file format with memory mapping, caching to increase speed, and quantization.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Note that hanzi that are not grouped into dictionary words (and are not identified as single\u00c2\u00ad hanzi words), or into one of the other categories of words discussed in this paper, are left unattached and tagged as unknown words.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Further, the probing hash table does only one random lookup per query, explaining why it is faster on large data.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Table 4 Differences in performance between our system and Wang, Li, and Chang (1992).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In the following, we use the notation wi wj to mean that (wi, r, wj) E A; r we also use wi wj to denote an arc with unspecified label and wi \u2014*\u2217 wj for the reflexive and transitive closure of the (unlabeled) arc relation.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"When the connective is an adverbial, there is much less clarity as to the range of the spans.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"These make left-to-right query patterns convenient, as the application need only provide a state and the word to append, then use the returned state to append another word, etc.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Maamouri et al.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"When SRILM estimates a model, it sometimes removes n-grams but not n + 1-grams that extend it to the left.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The goal of machine translation is the translation of a text given in some source language into a target language.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The annotator can then \u00e2\u0080\u009cclick away\u00e2\u0080\u009d those words that are here not used as connectives (such as the conjunction und (\u00e2\u0080\u0098and\u00e2\u0080\u0099) used in lists, or many adverbials that are ambiguous between connective and discourse particle).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"5 We choose these two metrics over the Variation Information measure due to the deficiencies discussed in Gao and Johnson (2008).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"\u00e2\u0080\u00a2 Some tools would allow for the desired annotation mode, but are so complicated (they can be used for many other purposes as well) that annotators take a long time getting used to them.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"First, we describe how the caseframes are represented and learned.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Finally, we add \u00e2\u0080\u009cDT\u00e2\u0080\u009d to the tags for definite nouns and adjectives (Kulick et al., 2006).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Chris Dyer integrated the code into cdec.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Then, token- level HMM emission parameters are drawn conditioned on these assignments such that each word is only allowed probability mass on a single assigned tag.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Hyperparameter settings are sorted according to the median one-to-one metric over runs.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We have provided methods for handling certain classes of unknown words, and models for other classes could be provided, as we have noted.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We call these N \u2212 1 words state.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Linear probing hash tables must have more buckets than entries, or else an empty bucket will never be found.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The paper explains the design decisions taken in the annotations, and describes a number of applications using this corpus with its multi-layer annotation.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The authors provided us with a ratio between TPT and SRI under different conditions. aLossy compression with the same weights. bLossy compression with retuned weights. ditions make the value appropriate for estimating repeated run times, such as in parameter tuning.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"What both of these approaches presume is that there is a sin\u00c2\u00ad gle correct segmentation for a sentence, against which an automatic algorithm can be compared.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Sc opi ng filters candidate if outside the anaphor\u00e2\u0080\u0099s scope.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In this case, this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Any NLP application that presumes as input unrestricted text requires an initial phase of text analysis; such applications involve problems as diverse as machine translation, information retrieval, and text-to-speech synthesis (TIS).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Given a sorted array A, these other packages use binary search to find keys in O(log |A|) time.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The Stanford parser includes both the manually annotated grammar (\u00c2\u00a74) and an Arabic unknown word model with the following lexical features: 1.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Each extraction pattern represents a linguistic expression and a syntactic position indicating where a role filler can be found.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"A novel element of our model is the ability to capture type-level tag frequencies.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The interdependence between fb or 1\/!i, and 5:2 is not captured by our model, but this could easily be remedied.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"An example is in (i), where the system fails to group t;,f;?\"$?t!: lin2yang2gang3 as a name, because all three hanzi can in principle be separate words (t;,f; lin2 'wood';?\"$ yang2 'ocean'; ?t!; gang3 'harbor').","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Then, the German infinitive 'besuchen' and the negation particle 'nicht' are translated.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The set is then compared with the set generated from the Penn Treebank parse to determine the precision and recall.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"A totally non\u00c2\u00ad stochastic rule-based system such as Wang, Li, and Chang's will generally succeed in such cases, but of course runs the risk of overgeneration wherever the single-hanzi word is really intended.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We can show that languages generated by LCFRS's are semilinear as long as the composition operation does not remove any terminal symbols from its arguments.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"What both of these approaches presume is that there is a sin\u00c2\u00ad gle correct segmentation for a sentence, against which an automatic algorithm can be compared.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Let H be the set of hanzi, p be the set of pinyin syllables with tone marks, and P be the set of grammatical part-of-speech labels.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Both BBN and NYU have tagged their own data to supplement the official training data.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We will refer to the semantic classes that co-occur with a caseframe as the semantic expectations of the caseframe.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The basic strategy is, for a given pair of entity types, to start with some examples, like several famous book title and author pairs; and find expressions which contains those names; then using the found expressions, find more author and book title pairs.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The details are given in (Tillmann, 2000).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"There are simply not enough votes remaining to allow any of the crossing structures to enter the hypothesized constituent set.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The accuracy results for segmentation, tagging and parsing using our different models and our standard data split are summarized in Table 1.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For RandLM and IRSTLM, the effect of caching can be seen on speed and memory usage.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In this paper, we have presented a new, e\u00c3\u0086cient DP-based search procedure for statistical machine translation.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The 4th block contains instance-weighting models trained on all features, used within a MAP TM combination, and with a linear LM mixture.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"It is important to bear in mind, though, that this is not an inherent limitation of the model.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The first concerns how to deal with ambiguities in segmentation.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In this paper, we show that the maximum entropy framework is able to make use of global information directly, and achieves performance that is comparable to the best previous machine learning-based NERs on MUC6 and MUC7 test data.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"During coreference resolution, BABAR checks (1) whether the anaphor is among the lexical expectations for the caseframe that extracts the candidate antecedent, and (2) whether the candidate is among the lexical expectations for the caseframe that extracts the anaphor.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"One striking example is the error reduction for Spanish, which reduces error by 36.5% and 24.7% for the best and median settings respectively.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In a fully supervised setting, the task is to learn a function f such that for all i = 1...m, f (xi,i, 12,i) = yz.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"2.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"TIS systems in general need to do more than simply compute the.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Each visited entry wni stores backoff b(wni ).","label":0},{"source_text":"Here we present two algorithms.","target_text":"The following features were used: full-string=x The full string (e.g., for Maury Cooper, full- s tring=Maury_Cooper). contains(x) If the spelling contains more than one word, this feature applies for any words that the string contains (e.g., Maury Cooper contributes two such features, contains (Maury) and contains (Cooper) . allcapl This feature appears if the spelling is a single word which is all capitals (e.g., IBM would contribute this feature). allcap2 This feature appears if the spelling is a single word which is all capitals or full periods, and contains at least one period.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Under this scheme, n human judges are asked independently to segment a text.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"It is closer to the smaller value of precision and recall when there is a large skew in their values.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The type of alignment we have considered so far requires the same length for source and target sentence, i.e. I = J. Evidently, this is an unrealistic assumption, therefore we extend the concept of inverted alignments as follows: When adding a new position to the coverage set C, we might generate either \u00c3\u0086 = 0 or \u00c3\u0086 = 1 new target words.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"BABAR achieved recall in the 4250% range for both domains, with 76% precision overall for terrorism and 87% precision for natural disasters.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Section 4.1 explained that state s is stored by applications with partial hypotheses to determine when they can be recombined.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In Section 3, we introduce our novel concept to word reordering and a DP-based search, which is especially suitable for the translation direction from German to English.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"To evaluate proper-name identification, we randomly se\u00c2\u00ad lected 186 sentences containing 12,000 hanzi from our test corpus and segmented the text automatically, tagging personal names; note that for names, there is always a sin\u00c2\u00ad gle unambiguous answer, unlike the more general question of which segmentation is correct.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Step 4.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We used the TF\/ITF metric to identify keywords.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This has solutions: where pI(s|t) is derived from the IN corpus using relative-frequency estimates, and po(s|t) is an instance-weighted model derived from the OUT corpus.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Unfortunately, we were unable to correctly run the IRSTLM quantized variant.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This is a straightforward technique that is arguably better suited to the adaptation task than the standard method of treating representative IN sentences as queries, then pooling the match results.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As a first step towards modeling transliterated names, we have collected all hanzi occurring more than once in the roughly 750 foreign names in our dictionary, and we estimate the probabil\u00c2\u00ad ity of occurrence of each hanzi in a transliteration (pTN(hanzi;)) using the maximum likelihood estimate.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"A different but supplementary perspective on discourse-based information structure is taken 11ventionalized patterns (e.g., order of informa by one of our partner projects, which is inter tion in news reports).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"We respond to this on the one hand with a format for its underspecification (see 2.4) and on the other hand with an additional level of annotation that attends only to connectives and their scopes (see 2.5), which is intended as an intermediate step on the long road towards a systematic and objective treatment of rhetorical structure.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We utilized two kinds of datasets in our experiments: (i) monolingual treebanks9 and (ii) large amounts of parallel text with English on one side.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Table 5: Individual Performance of KSs for Disasters (e.g., \u00e2\u0080\u009cthe mayor\u00e2\u0080\u009d vs. \u00e2\u0080\u009cthe journalist\u00e2\u0080\u009d).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The tokens w are generated by token-level tags t from an HMM parameterized by the lexicon structure.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"However, our full model takes advantage of word features not present in Grac\u00c2\u00b8a et al.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The same argument holds for resolving PP attachment of a prefixed preposition or marking conjunction of elements of any kind.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"shortest match at each point.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Hyperparameters Our model has two Dirichlet concentration hyperparameters: \u00ce\u00b1 is the shared hyperparameter for the token-level HMM emission and transition distributions.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"It is sometimes claimed that one of the advantages of dependency grammar over approaches based on constituency is that it allows a more adequate treatment of languages with variable word order, where discontinuous syntactic constructions are more common than in languages like English (Mel\u2019\u02c7cuk, 1988; Covington, 1990).","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"(Webber et al., 2003)).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Clearly, explicitly modeling such a powerful constraint on tagging assignment has a potential to significantly improve the accuracy of an unsupervised part-of-speech tagger learned without a tagging dictionary.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The work of Rounds (1969) shows that the path sets of trees derived by IG's (like those of TAG's) are context-free languages.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"What both of these approaches presume is that there is a sin\u00c2\u00ad gle correct segmentation for a sentence, against which an automatic algorithm can be compared.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: The government said it ...","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Now assume we have n pairs (xi,, x2,i) drawn from X1 X X2, where the first m pairs have labels whereas for i = m+ 1...n the pairs are unlabeled.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"our full model yields 39.3% average error reduction across languages when compared to the basic configuration (1TW).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"In Modern Hebrew (Hebrew), a Semitic language with very rich morphology, particles marking conjunctions, prepositions, complementizers and relativizers are bound elements prefixed to the word (Glinert, 1989).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In German, the verbgroup usually consists of a left and a right verbal brace, whereas in English the words of the verbgroup usually form a sequence of consecutive words.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We define a symmetric similarity function K(uZ7 uj) over two foreign language vertices uZ7 uj E Vf based on the co-occurrence statistics of the nine feature concepts given in Table 1.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We can give a tree pumping lemma for TAG's by adapting the uvwxy-theorem for CFL's since the tree sets of TAG's have independent and context-free paths.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"An easy way to achieve this is to put the domain-specific LMs and TMs into the top-level log-linear model and learn optimal weights with MERT (Och, 2003).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The use of the Good-Turing equation presumes suitable estimates of the unknown expectations it requires.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Ex: The regime gives itself the right...","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"32 81.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Clearly, retaining the original frequencies is important for good performance, and globally smoothing the final weighted frequencies is crucial.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Evaluation of Morphological Analysis.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"7 68.3 56.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For each set, the phrases with bracketed frequencies are considered not paraphrases in the set.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In defining LCFRS's, we hope to generalize the definition of CFG's to formalisms manipulating any structure, e.g. strings, trees, or graphs.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Intuitively, it places more weight on OUT when less evidence from IN is available.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"On each language we investigate the contribution of each component of our model.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"3.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In the named entity domain these rules were Each of these rules was given a strength of 0.9999.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"4 69.0 51.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We thank United Informatics for providing us with our corpus of Chinese text, and BDC for the 'Behavior ChineseEnglish Electronic Dictionary.'","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"However, we needed to restrict ourselves to these languages in order to be able to evaluate the performance of our approach.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Later, BerkeleyLM (Pauls and Klein, 2011) described ideas similar to ours.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"rently, some annotations (in particular the connectives and scopes) have already moved beyond the core corpus; the others will grow step by step.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Memory usage is likely much lower than ours. fThe original paper (Germann et al., 2009) provided only 2s of query timing and compared with SRI when it exceeded available RAM.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Note that the sets of possible classifiers for a given noun can easily be encoded on that noun by grammatical features, which can be referred to by finite-state grammatical rules.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In that work, mutual information was used to decide whether to group adjacent hanzi into two-hanzi words.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"AdaBoost.MH maintains a distribution over instances and labels; in addition, each weak-hypothesis outputs a confidence vector with one confidence value for each possible label.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The tree denoting this derivation of 7 is rooted with a node labeled 7' having k subtrees for the derivations of 71, ... ,7a.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We use two different similarity functions to define the edge weights among the foreign vertices and between vertices from different languages.","label":0},{"source_text":"Here we present two algorithms.","target_text":"N, portion of examples on which both classifiers give a label rather than abstaining), and the proportion of these examples on which the two classifiers agree.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"These knowledge sources determine whether the contexts surrounding an anaphor and antecedent are compatible.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The first step in the learning process is to generate training examples consisting of anaphor\/antecedent resolutions.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The code is opensource, has minimal dependencies, and offers both C++ and Java interfaces for integration.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Interpolation search has a more expensive pivot but performs less pivoting and reads, so it is slow on small data and faster on large data.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"(7) \u03c6 s,t This is a somewhat less direct objective than used by Matsoukas et al, who make an iterative approximation to expected TER.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Step 1.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The development of a na\u00efve Bayes classifier involves learning how much each parser should be trusted for the decisions it makes.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The extent to which this constraint is enforced varies greatly across existing methods.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Acronyms (ACRO): Words made up of all capitalized letters in the text zone will be stored as acronyms (e.g., IBM).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"3 The Coreference Resolution Model.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In principle, it would be possible to encode the exact position of the syntactic head in the label of the arc from the linear head, but this would give a potentially infinite set of arc labels and would make the training of the parser very hard.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Thus we are interested not in extraction, but actual generation from representations that may be developed to different degrees of granularity.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The use of global features has improved the performance on MUC6 test data from 90.75% to 93.27% (27% reduction in errors), and the performance on MUC7 test data from 85.22% to 87.24% (14% reduction in errors).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Our original hope in combining these parsers is that their errors are independently distributed.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For all variants, we found that BerkeleyLM always rounds the floating-point mantissa to 12 bits then stores indices to unique rounded floats.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Because many foreign word types are not aligned to an English word (see Table 3), and we do not run label propagation on the foreign side, we expect the projected information to have less coverage.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In a fully supervised setting, the task is to learn a function f such that for all i = 1...m, f (xi,i, 12,i) = yz.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The availability of comparable corpora is limited, which is a significant limitation on the approach.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Not surprisingly some semantic classes are better for names than others: in our corpora, many names are picked from the GRASS class but very few from the SICKNESS class.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The algorithm works due to the fact that not all permutations of cities have to be considered explicitly.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The same numbers were used for each data structure.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"As a result, Habash et al.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In this paper, we will propose an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"However, those methods need initial seeds, so the relation between entities has to be known in advance.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The combining algorithm is presented with the candidate parses and asked to choose which one is best.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We propose an unsupervised method to discover paraphrases from a large untagged corpus, without requiring any seed phrase or other cue.","label":1},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Ablation Analysis We evaluate the impact of incorporating various linguistic features into our model in Table 3.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We refer to (T , W ) as the lexicon of a language and \u00cf\u0088 for the parameters for their generation; \u00cf\u0088 depends on a single hyperparameter \u00ce\u00b2.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"For the inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J), we drop the dependence on the target sentence length I. 2.2 Word Joining.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The foreign language vertices (denoted by Vf) correspond to foreign trigram types, exactly as in Subramanya et al. (2010).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Reading the following record\u2019s offset indicates where the block ends.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"It is not easy to make a clear definition of \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The equivalence of CC's with this restriction to TAG's and HG's is, however, still an open problem.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"A totally non\u00c2\u00ad stochastic rule-based system such as Wang, Li, and Chang's will generally succeed in such cases, but of course runs the risk of overgeneration wherever the single-hanzi word is really intended.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In these experiments, the input lacks segmentation markers, hence the slightly different dev set baseline than in Table 6.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"For comparison to information-retrieval inspired baselines, eg (L\u00a8u et al., 2007), we select sentences from OUT using language model perplexities from IN.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"However, MADA is language-specific and relies on manually constructed dictionaries.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Even without features, but still using the tag prior, our median result is 52.0%, still significantly outperforming Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"While the linear precedence of segmental morphemes within a token is subject to constraints, the dominance relations among their mother and sister constituents is rather free.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Call the crossing constituents A and B.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"2.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The use of the Good-Turing equation presumes suitable estimates of the unknown expectations it requires.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"ments contained 322 anaphoric links.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Step 3.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"i=1 (f,v)\u00e2\u0088\u0088Wi","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"We have used the Java-based opennlp maximum entropy package1.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We carried out translation experiments in two different settings.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The same numbers were used for each data structure.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The final strong hypothesis, denoted 1(x), is then the sign of a weighted sum of the weak hypotheses, 1(x) = sign (Vii atht(x)), where the weights at are determined during the run of the algorithm, as we describe below.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Taking only the highest frequency rules is much &quot;safer&quot;, as they tend to be very accurate.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"As can be seen in the example, the first two phrases have a different order of NE names from the last two, so we can determine that the last two phrases represent a reversed relation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Table 3 Classes of words found by ST for the test corpus.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"5.1 Parsing Models.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"This limits the number of NE category pairs to 2,000 and the number of NE pair instances to 0.63 million.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"of Articles No.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Except our own and MENE + reference resolution, the results in Table 6 are all official MUC7 results.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We of course also fail to identify, by the methods just described, given names used without their associated family name.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Juri Ganitkevitch answered questions about Joshua.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We will return to these issues in the discussion section.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The cost estimate, cost(i\u00c2\u00a5JJ1l.fn is computed in the obvious way by summing the negative log probabilities of i\u00c2\u00a5JJ1l.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"1 1 0.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Let us consider an example of ambiguity caused by devocalization.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The corpus was wordaligned using both HMM and IBM2 models, and the phrase table was the union of phrases extracted from these separate alignments, with a length limit of 7.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Many morphological decisions are based on long distance dependencies, and when the global syntactic evidence disagrees with evidence based on local linear context, the two models compete with one another, despite the fact that the PCFG takes also local context into account.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For a description of the application of AdaBoost to various NLP problems see the paper by Abney, Schapire, and Singer in this volume.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In the third and final scheme, denoted Path, we keep the extra infor2Note that this is a baseline for the parsing experiment only (Experiment 2).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We will describe the evaluation of such clusters in the next subsection.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"As lower frequency examples include noise, we set a threshold that an NE category pair should appear at least 5 times to be considered and an NE instance pair should appear at least twice to be considered.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In this section, we try to compare our results with those obtained by IdentiFinder ' 97 (Bikel et al., 1997), IdentiFinder ' 99 (Bikel et al., 1999), and MENE (Borthwick, 1999).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The DL-CoTrain algorithm can be motivated as being a greedy method of satisfying the above 2 constraints.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Kollege.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We have shown that, at least given independent human judgments, this is not the case, and that therefore such simplistic measures should be mistrusted.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"This feature has a linguistic justification.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"For all grammars, we use fine-grained PoS tags indicating various morphological features annotated therein.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"(2009), who also incorporate a sparsity constraint, but does via altering the model objective using posterior regularization.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Both of these analyses are shown in Figure 4; fortunately, the correct analysis is also the one with the lowest cost, so it is this analysis that is chosen.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In the initial release of the ATB, inter-annotator agreement was inferior to other LDC treebanks (Maamouri et al., 2008).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"A different knowledge source, called CFSemCFSem, compares the semantic expectations of the caseframe that extracts the anaphor with the semantic expectations of the caseframe that extracts the candidate.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The developers aimed to reduce memory consumption at the expense of time.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"com \u00c2\u00a7Cambridge, UK Email: nc201@eng.cam.ac.uk \u00c2\u00a9 1996 Association for Computational Linguistics (a) B ) ( , : & ; ? ' H o w d o y o u s a y o c t o p u s i n J a p a n e s e ? ' (b) P l a u s i b l e S e g m e n t a t i o n I B X I I 1 : & I 0 0 r i 4 w e n 2 z h a n g l y u 2 z e n 3 m e 0 s h u o l ' J a p a n e s e ' ' o c t o p u s ' ' h o w ' ' s a y ' (c) Figure 1 I m p l a u s i b l e S e g m e n t a t i o n [\u00c2\u00a7] lxI 1:&I ri4 wen2 zhangl yu2zen3 me0 shuol 'Japan' 'essay' 'fish' 'how' 'say' A Chinese sentence in (a) illustrating the lack of word boundaries.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The portions of information in the large window can be individually clicked visible or invisible; here we have chosen to see (from top to bottom) \u00e2\u0080\u00a2 the full text, \u00e2\u0080\u00a2 the annotation values for the activated annotation set (co-reference), \u00e2\u0080\u00a2 the actual annotation tiers, and \u00e2\u0080\u00a2 the portion of text currently \u00e2\u0080\u0098in focus\u00e2\u0080\u0099 (which also appears underlined in the full text).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"When finished, the whole material is written into an XML-structured annotation file.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In this section, we describe how contextual role knowledge is represented and learned.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Note that in line 4 the last visited position for the successor hypothesis must be m. Otherwise , there will be four uncovered positions for the predecessor hypothesis violating the restriction.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For E(ni1s), then, we substitute a smooth S against the number of class elements.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Past work however, has typically associ n = n P (Ti)P (Wi|Ti) = i=1 1 n K n ated these features with token occurrences, typically in an HMM.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"64 76.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We have not yet tried this.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The annotator can then \u00e2\u0080\u009cclick away\u00e2\u0080\u009d those words that are here not used as connectives (such as the conjunction und (\u00e2\u0080\u0098and\u00e2\u0080\u0099) used in lists, or many adverbials that are ambiguous between connective and discourse particle).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For each label (Per s on, organization and Location), take the n contextual rules with the highest value of Count' (x) whose unsmoothed3 strength is above some threshold pmin.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Having defined LCFRS's, in Section 4.2 we established the semilinearity (and hence constant growth property) of the languages generated.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In our model, however, all lattice paths are taken to be a-priori equally likely.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"There exist a few robust broad-coverage parsers that produce non-projective dependency structures, notably Tapanainen and J\u00a8arvinen (1997) and Wang and Harper (2004) for English, Foth et al. (2004) for German, and Holan (2004) for Czech.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"A Single Generative Model for Joint Morphological Segmentation and Syntactic Parsing","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"11 taTweel (-) is an elongation character used in Arabic script to justify text.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"For other languages, we use the CoNLL-X multilingual dependency parsing shared task corpora (Buchholz and Marsi, 2006) which include gold POS tags (used for evaluation).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"They also describe an application of cotraining to classifying web pages (the to feature sets are the words on the page, and other pages pointing to the page).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Our experimental setup therefore is designed to serve two goals.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This suggests that the backoff model is as reasonable a model as we can use in the absence of further information about the expected cost of a plural form.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"When two partial hypotheses have equal state (including that of other features), they can be recombined and thereafter efficiently handled as a single packed hypothesis.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Because b is a function, no additional hypothesis splitting happens.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Se ma nti c (a) filters candidate if its semantic tags d o n \u00e2\u0080\u0099 t i n t e r s e c t w i t h t h o s e o f t h e a n a p h o r .","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Note that the backoff model assumes that there is a positive correlation between the frequency of a singular noun and its plural.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"This allows an unbounded amount of information about two separate paths (e.g. an encoding of their length) to be combined and used to influence the later derivation.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Compared to the EMEA\/EP setting, the two domains in the NIST setting are less homogeneous and more similar to each other; there is also considerably more IN text available.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We considered using the MUC6 and MUC7 data sets, but their training sets were far too small to learn reliable co-occurrence statistics for a large set of contextual role relationships.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"3 The Coreference Resolution Model.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Table 5: Effect of the beam threshold on the number of search errors (147 sentences).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Second, we treat the projected labels as features in an unsupervised model (\u00a75), rather than using them directly for supervised training.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We again adopt an approach where we alternate between two classifiers: one classifier is modified while the other remains fixed.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Under this scheme, n human judges are asked independently to segment a text.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"First, in section 4, we evaluate the graph transformation techniques in themselves, with data from the Prague Dependency Treebank and the Danish Dependency Treebank.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This class-based model gives reasonable results: for six radical classes, Table 1 gives the estimated cost for an unseen hanzi in the class occurring as the second hanzi in a double GIVEN name.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Dropping the conditioning on 0 for brevity, and letting \u00afc\u03bb(s, t) = c\u03bb(s, t) + yu(s|t), and \u00afc\u03bb(t) = 4Note that the probabilities in (7) need only be evaluated over the support of \u02dcp(s, t), which is quite small when this distribution is derived from a dev set.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The weak learner for two-class problems computes a weak hypothesis h from the input space into the reals (h : 2x -4 R), where the sign4 of h(x) is interpreted as the predicted label and the magnitude I h(x)I is the confidence in the prediction: large numbers for I h(x)I indicate high confidence in the prediction, and numbers close to zero indicate low confidence.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In our experiments we set the parameter values randomly, and then ran EM to convergence.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For definite NPs, the results are a mixed bag: some knowledge sources increased recall a little, but at the expense of some precision.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Our monolingual similarity function (for connecting pairs of foreign trigram types) is the same as the one used by Subramanya et al. (2010).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Consider first the examples in (2).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We will return to these issues in the discussion section.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The first step in the learning process is to generate training examples consisting of anaphor\/antecedent resolutions.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"1).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"To provide a thorough analysis, we evaluated three baselines and two oracles in addition to two variants of our graph-based approach.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Frontier nodes are annotated by zero arty functions corresponding to elementary structures.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Smith estimates Lotus will make a profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d, our system extracts \u00e2\u0080\u009cSmith esti mates Lotus\u00e2\u0080\u009d as an instance.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The knowledge base then can be tested for its relation-inference capabilities on the basis of full-blown co-reference information.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Each set is assigned two values: belief and plausibility.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The overall performance of our joint framework demonstrates that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperforms upper bounds proposed by previous joint disambiguation systems and achieves segmentation and parsing results on a par with state-of-the-art standalone applications results.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We simulate lexical constraints by using an external lexical resource against which we verify whether OOV segments are in fact valid Hebrew lexemes.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The relativizer f(\u201cthat\u201d) for example, may attach to an arbitrarily long relative clause that goes beyond token boundaries.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"First, the model assumes independence between the first and second hanzi of a double given name.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"For the TM, this is: where cI(s, t) is the count in the IN phrase table of pair (s, t), po(s|t) is its probability under the OUT TM, and cI(t) = &quot;s, cI(s', t).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"At most one feature in this group will be set to 1.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Second, we identified the 100 most frequent nouns in the training corpus and manually labeled them with semantic tags.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Each xii is a member of X, where X is a set of possible features.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Part-of-speech (POS) tag distributions are known to exhibit sparsity \u00e2\u0080\u0094 a word is likely to take a single predominant tag in a corpus.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Case and Zone: If the token starts with a capital letter (initCaps), then an additional feature (init- Caps, zone) is set to 1.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"From the point of view of computational implementation this can be problematic, since the inclusion of non-projective structures makes the parsing problem more complex and therefore compromises efficiency and in practice also accuracy and robustness.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"nan2gual 'pumpkin.'","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Given that part-of-speech labels are properties of words rather than morphemes, it follows that one cannot do part-of-speech assignment without having access to word-boundary information.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This range is collapsed to a number of buckets, typically by taking the hash modulo the number of buckets.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"These three parsers have given the best reported parsing results on the Penn Treebank Wall Street Journal corpus (Marcus et al., 1993).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The CoBoost algorithm described above divides the function Zco into two parts: Zco = 40 + 40.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This task measures how well each package performs in machine translation.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We have shown the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","label":1},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We believe that the underlying principles of the maximum entropy framework are suitable for exploiting information from diverse sources.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The first author is supported by a National Defense Science and Engineering Graduate (NDSEG) fellowship.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We quantify error categories in both evaluation settings.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Exposing this information to the decoder will lead to better hypothesis recombination.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Since the early days of statistical NLP, researchers have observed that a part-of-speech tag distribution exhibits \u00e2\u0080\u009cone tag per discourse\u00e2\u0080\u009d sparsity \u00e2\u0080\u0094 words are likely to select a single predominant tag in a corpus, even when several tags are possible.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"However, the next step is clearly different.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"computing the precision of the other's judgments relative to this standard.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"In a model we built with default settings, 1.2% of n + 1-grams were missing their ngram suffix.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"This limitation is the obstacle to making the technology \u00e2\u0080\u009copen domain\u00e2\u0080\u009d.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Hyperparameter settings are sorted according to the median one-to-one metric over runs.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Following the setup of Johnson (2007), we use the whole of the Penn Treebank corpus for training and evaluation on English.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"What both of these approaches presume is that there is a sin\u00c2\u00ad gle correct segmentation for a sentence, against which an automatic algorithm can be compared.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.1 Reliable Case Resolutions.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Figure 4 Input lattice (top) and two segmentations (bottom) of the sentence 'How do you say octopus in Japanese?'.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"49 99.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Since pronouns carry little semantics of their own, resolving them depends almost entirely on context.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Informally, two or more paths can be dependent on each other: for example, they could be required to be of equal length as in the trees in Figure 4. generates such a tree set.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The ATB gives several different analyses to these words to indicate different types of coordination.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The model we use provides a simple framework in which to incorporate a wide variety of lexical information in a uniform way.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Morphological segmentation decisions in our model are delegated to a lexeme-based PCFG and we show that using a simple treebank grammar, a data-driven lexicon, and a linguistically motivated unknown-tokens handling our model outperforms (Tsarfaty, 2006) and (Cohen and Smith, 2007) on the joint task and achieves state-of-the-art results on a par with current respective standalone models.2","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The entries in this table can be compared with those of Table 3 to see how the performance of the combining techniques degrades in the presence of an inferior parser.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"It may be more realistic to replace the second criteria with a softer one, for example (Blum and Mitchell 98) suggest the alternative Alternatively, if Ii and 12 are probabilistic learners, it might make sense to encode the second constraint as one of minimizing some measure of the distance between the distributions given by the two learners.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We checked whether the discovered links are listed in WordNet.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"When a token fmnh is to be interpreted as the lexeme sequence f\/REL mnh\/VB, the analysis introduces two distinct entities, the relativizer f (\u201cthat\u201d) and the verb mnh (\u201ccounted\u201d), and not as the complex entity \u201cthat counted\u201d.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Several papers report the use of part-of-speech information to rank segmentations (Lin, Chiang, and Su 1993; Peng and Chang 1993; Chang and Chen 1993); typically, the probability of a segmentation is multiplied by the probability of the tagging(s) for that segmentation to yield an estimate of the total probability for the analysis.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"For each experiment we gave an nonparametric and a parametric technique for combining parsers.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"annotation guidelines that tell annotators what to do in case of doubt.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The second setting uses the news-related subcorpora for the NIST09 MT Chinese to English evaluation8 as IN, and the remaining NIST parallel Chinese\/English corpora (UN, Hong Kong Laws, and Hong Kong Hansard) as OUT.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"ICOC and CSPP contributed the greatest im provements.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Except our own and MENE + reference resolution, the results in Table 6 are all official MUC7 results.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The frequency of the Company \u00e2\u0080\u0093 Company domain ranks 11th with 35,567 examples.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The goal of our research was to explore the use of contextual role knowledge for coreference resolution.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Memory usage in PROBING is high, though SRILM is even larger, so where memory is of concern we recommend using TRIE, if it fits in memory.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In such cases we assign all of the estimated probability mass to the form with the most likely pronunciation (determined by inspection), and assign a very small probability (a very high cost, arbitrarily chosen to be 40) to all other variants.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Adam Pauls provided a pre-release comparison to BerkeleyLM and an initial Java interface.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We extend the Matsoukas et al approach in several ways.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The data is sorted based on the frequency of the context (\u00e2\u0080\u009ca unit of\u00e2\u0080\u009d appeared 314 times in the corpus) and the NE pair instances appearing with that context are shown with their frequency (e.g. \u00e2\u0080\u009cNBC\u00e2\u0080\u009d and \u00e2\u0080\u009cGeneral Electric Co.\u00e2\u0080\u009d appeared 10 times with the context \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A list of words occurring more than 10 times in the training data is also collected (commonWords).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"4.3 Morphological Analysis.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"41.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In the natural disasters domain, agents are often forces of nature, such as hurricanes or wildfires.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The method shares some characteristics of the decision list algorithm presented in this paper.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For each token , zero, one, or more of the features in each feature group are set to 1.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"When this metric is less than 0.5, we expect to incur more errors' than we will remove by adding those constituents to the parse.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"On the other hand, only very restricted reorderings are necessary, e.g. for the translation direction from Table 2: Coverage set hypothesis extensions for the IBM reordering.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The computing time, the number of search errors, and the multi-reference WER (mWER) are shown as a function of t0.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Combining multiple highly-accurate independent parsers yields promising results.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"State will ultimately be used as context in a subsequent query.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Although we have not done any experiments on other languages, this way of using global features from a whole document should be applicable to other languages.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(6), with W+ > W_.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We indicate whether a context with zero log backoff will extend using the sign bit: +0.0 for contexts that extend and \u22120.0 for contexts that do not extend.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Adding the isolated constituents to our hypothesis parse could increase our expected recall, but in the cases we investigated it would invariably hurt our precision more than we would gain on recall.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"An inverted alignment is defined as follows: inverted alignment: i ! j = bi: Target positions i are mapped to source positions bi.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We were intentionally lenient with our baselines: bilingual information by projecting POS tags directly across alignments in the parallel data.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Such tag sequences are often treated as \u201ccomplex tags\u201d (e.g.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Finally, we wish to reiterate an important point.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"9 50.2 +P RI OR be st me dia n 47.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The accuracies for link were 73% and 86% on two evaluated domains.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"11 taTweel (-) is an elongation character used in Arabic script to justify text.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Recent work has made significant progress on unsupervised POS tagging (Me\u00c2\u00b4rialdo, 1994; Smith and Eisner, 2005; Haghighi and Klein, 2006; Johnson,2007; Goldwater and Griffiths, 2007; Gao and John son, 2008; Ravi and Knight, 2009).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This is similar to using the Linux MAP POPULATE flag that is our default loading mechanism.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The data is sorted based on the frequency of the context (\u00e2\u0080\u009ca unit of\u00e2\u0080\u009d appeared 314 times in the corpus) and the NE pair instances appearing with that context are shown with their frequency (e.g. \u00e2\u0080\u009cNBC\u00e2\u0080\u009d and \u00e2\u0080\u009cGeneral Electric Co.\u00e2\u0080\u009d appeared 10 times with the context \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The original tag set for the CoNLL-X Dutch data set consists of compounded tags that are used to tag multi-word units (MWUs) resulting in a tag set of over 300 tags.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"87 Table 7: Test set results.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We develop our POS induction model based on the feature-based HMM of Berg-Kirkpatrick et al. (2010).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The breakdown of the different types of words found by ST in the test corpus is given in Table 3.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"41.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The pseudo-code describing the algorithm is given in Fig.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Preprocessing the raw trees improves parsing performance considerably.9 We first discard all trees dominated by X, which indicates errors and non-linguistic text.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"72 78.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"They cluster NE instance pairs based on the words in the contexts using a bag- of-words method.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The government has to make a decision, and do it quickly.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This has the effect of randomly permuting vocabulary identifiers, meeting the requirements of interpolation search when vocabulary identifiers are used as keys.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Finally, the focus\/background partition is annotated, together with the focus question that elicits the corresponding answer.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Since our graph is built from a parallel corpus, we can use standard word alignment techniques to align the English sentences De 5Note that many combinations are impossible giving a PMI value of 0; e.g., when the trigram type and the feature instantiation don\u2019t have words in common. and their foreign language translations Df.6 Label propagation in the graph will provide coverage and high recall, and we therefore extract only intersected high-confidence (> 0.9) alignments De\ufffdf.","label":0},{"source_text":"There is no global pruning.","target_text":"diesem 3.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The use of ILP in learning the desired grammar significantly increases the computational complexity of this method.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"1 The apparent difficulty of adapting constituency models to non-configurational languages has been one motivation for dependency representations (Hajic\u00cb\u0087 and Zema\u00c2\u00b4nek, 2004; Habash and Roth, 2009).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We incorporate instance weighting into a mixture-model framework, and find that it yields consistent improvements over a wide range of baselines.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"It reads a file with a list of German connectives, and when a text is opened for annotation, it highlights all the words that show up in this list; these will be all the potential connectives.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Let s = a + b.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Here NO counts the number of hypothesized constituents in the development set that match the binary predicate specified as an argument.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In considering recognition of LCFRS's, we make further assumption concerning the contribution of each structure to the input string, and how the composition operations combine structures in this respect.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Both parameters depend on a single hyperparameter \u00ce\u00b1.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"It has no syntactic function.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"10 Chinese speakers may object to this form, since the suffix f, menD (PL) is usually restricted to.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"BABAR uses a DempsterShafer decision model (Stefik, 1995) to combine the evidence provided by the knowledge sources.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(6), with W+ > W_.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"3 68.4 49.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The approach recursively evaluates a quantity Q(C; j), where C is the set of already visited cities and sj is the last visited city.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Across all languages, high performance can be attained by selecting a single tag per word type.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"When a collision occurs, linear probing places the entry to be inserted in the next (higher index) empty bucket, wrapping around as necessary.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The two knowledge sources that use semantic expectations, WordSemCFSem and CFSemCFSem, always return values of -1 or 0.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The cost of storing these averages, in bits, is Because there are comparatively few unigrams, we elected to store them byte-aligned and unquantized, making every query faster.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"99 94.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The two knowledge sources that use semantic expectations, WordSemCFSem and CFSemCFSem, always return values of -1 or 0.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In section 4 we evaluate these transformations with respect to projectivized dependency treebanks, and in section 5 they are applied to parser output.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Asking the annotator to also formulate the question is a way of arriving at more reproducible decisions.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Section 3 discusses the applications that have been completed with PCC, or are under way, or are planned for the future.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"29 \u00e2\u0080\u0094 95.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The F- measure score increased for both domains, reflecting a substantial increase in recall with a small decrease in precision.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(b) 89 :1 t& tal de cai2neng2 hen3 he DE talent very 'He has great talent' f.b ga ol hig h While the current algorithm correctly handles the (b) sentences, it fails to handle the (a) sentences, since it does not have enough information to know not to group the sequences.ma3lu4 and?]cai2neng2 respectively.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Global context from the whole document is available and can be exploited in a natural manner with a maximum entropy classifier.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"01 75.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"BABAR then computes statistics over the training examples measuring the frequency with which extraction patterns and noun phrases co-occur in coreference resolutions.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, in the sentence that starts with \u00e2\u0080\u009cBush put a freeze on . . .","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We identified three ways that contextual roles can be exploited: (1) by identifying caseframes that co-occur in resolutions, (2) by identifying nouns that co-occur with case- frames and using them to crosscheck anaphor\/candidate compatibility, (3) by identifying semantic classes that co- occur with caseframes and using them to crosscheck anaphor\/candidate compatability.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The first term in the objective function is the graph smoothness regularizer which encourages the distributions of similar vertices (large wij) to be similar.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"N, portion of examples on which both classifiers give a label rather than abstaining), and the proportion of these examples on which the two classifiers agree.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"ATB CTB6 Negra WSJ Trees 23449 28278 20602 43948 Word Typess 40972 45245 51272 46348 Tokens 738654 782541 355096 1046829 Tags 32 34 499 45 Phrasal Cats 22 26 325 27 Test OOV 16.8% 22.2% 30.5% 13.2% Per Sentence Table 4: Gross statistics for several different treebanks.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Yet we note that the better grammars without pruning outperform the poorer grammars using this technique, indicating that the syntactic context aids, to some extent, the disambiguation of unknown tokens.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The corpus was wordaligned using both HMM and IBM2 models, and the phrase table was the union of phrases extracted from these separate alignments, with a length limit of 7.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We would like to thank Eugene Charniak, Michael Collins, and Adwait Ratnaparkhi for enabling all of this research by providing us with their parsers and helpful comments.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The first unsupervised algorithm we describe is based on the decision list method from (Yarowsky 95).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Using a wide-coverage morphological analyzer based on (Itai et al., 2006) should cater for a better coverage, and incorporating lexical probabilities learned from a big (unannotated) corpus (cf.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This task measures how well each package performs in machine translation.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(7) is at 0 when: 1) Vi : sign(gi (xi)) = sign(g2 (xi)); 2) Ig3(xi)l oo; and 3) sign(gi (xi)) = yi for i = 1, , m. In fact, Zco provides a bound on the sum of the classification error of the labeled examples and the number of disagreements between the two classifiers on the unlabeled examples.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The unlabeled data gives many such &quot;hints&quot; that two features should predict the same label, and these hints turn out to be surprisingly useful when building a classifier.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Table 2 shows single-threaded results, mostly for comparison to IRSTLM, and Table 3 shows multi-threaded results.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Now, for this application one might be tempted to simply bypass the segmentation problem and pronounce the text character-by-character.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This means that the PCC cannot grow particularly quickly.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"In general, the lemma of the previous section does not ensure that all the productions in the combined parse are found in the grammars of the member parsers.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The maximum likelihood estimates (i.e., parameter values which maximize 10) can not be found analytically, but the EM algorithm can be used to hill-climb to a local maximum of the likelihood function from some initial parameter settings.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"JI!","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This approach makes the training objective more complex by adding linear constraints proportional to the number of word types, which is rather prohibitive.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The number of NE instance pairs used in their experiment is less than half of our method.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We also mark all tags that dominate a word with the feminine ending :: taa mar buuTa (markFeminine).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"But we also need an estimate of the probability for a non-occurring though possible plural form like i\u00c2\u00a5JJ1l.f, nan2gua1-men0 'pumpkins.'","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We have not to date explored these various options.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"After the first step towards breadth had been taken with the PoS-tagging, RST annotation, and URML conversion of the entire corpus of 170 texts12 , emphasis shifted towards depth.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Given this similarity function, we define a nearest neighbor graph, where the edge weight for the n most similar vertices is set to the value of the similarity function and to 0 for all other vertices.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models (e.g., (Grosz et al., 1995; Grosz and Sidner, 1998)), syntactic algorithms (e.g., (Hobbs, 1978; Lappin and Le- ass, 1994)), and supervised machine learning systems (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Find keywords for each NE pair When we look at the contexts for each domain, we noticed that there is one or a few important words which indicate the relation between the NEs (for example, the word \u00e2\u0080\u009cunit\u00e2\u0080\u009d for the phrase \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We focus on phrases which connect two Named Entities (NEs), and proceed in two stages.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Automatic Paraphrase Discovery based on Context and Keywords between NE Pairs","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"As a result, Arabic sentences are usually long relative to English, especially after Length English (WSJ) Arabic (ATB) \u00e2\u0089\u00a4 20 41.9% 33.7% \u00e2\u0089\u00a4 40 92.4% 73.2% \u00e2\u0089\u00a4 63 99.7% 92.6% \u00e2\u0089\u00a4 70 99.9% 94.9% Table 2: Frequency distribution for sentence lengths in the WSJ (sections 2\u00e2\u0080\u009323) and the ATB (p1\u00e2\u0080\u00933).","label":0},{"source_text":"Here we present two algorithms.","target_text":"Zt can be written as follows Following the derivation of Schapire and Singer, providing that W+ > W_, Equ.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, there will remain a large number of words that are not readily adduced to any produc\u00c2\u00ad tive pattern and that would simply have to be added to the dictionary.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In Modern Hebrew (Hebrew), a Semitic language with very rich morphology, particles marking conjunctions, prepositions, complementizers and relativizers are bound elements prefixed to the word (Glinert, 1989).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In this paper, we offer broad insight into the underperformance of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","label":1},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Fortunately, there are only a few hundred hanzi that are particularly common in transliterations; indeed, the commonest ones, such as E. bal, m er3, and iij al are often clear indicators that a sequence of hanzi containing them is foreign: even a name like !:i*m xia4mi3-er3 'Shamir,' which is a legal Chi\u00c2\u00ad nese personal name, retains a foreign flavor because of liM.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Aside from adding a simple rule to correct alif deletion caused by the preposition J, no other language-specific processing is performed.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The composition operations are mapped onto operations that use concatenation to define the substrings spanned by the resulting structures.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Let H be the set of hanzi, p be the set of pinyin syllables with tone marks, and P be the set of grammatical part-of-speech labels.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"36.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The 4th block contains instance-weighting models trained on all features, used within a MAP TM combination, and with a linear LM mixture.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Note that in our formalism a weakhypothesis can abstain.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"On each step CoBoost searches for a feature and a weight so as to minimize either 40 or 40.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"If the parse contains productions from outside our grammar the machine has no direct method for handling them (e.g. the resulting database query may be syntactically malformed).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"On the English side, however, the vertices (denoted by Ve) correspond to word types.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The final score is obtained from: max e;e0 j2fJ\u00f4\u0080\u0080\u0080L;;Jg p($je; e0) Qe0 (e; I; f1; ; Jg; j); where p($je; e0) denotes the trigram language model, which predicts the sentence boundary $ at the end of the target sentence.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"However their work did not consider other types of lexical expectations (e.g., PP arguments), semantic expectations, or context comparisons like our case- frame network.(Niyu et al., 1998) used unsupervised learning to ac quire gender, number, and animacy information from resolutions produced by a statistical pronoun resolver.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Alternatively, h can be thought of as defining a decision list of rules x y ranked by their &quot;strength&quot; h(x, y).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"A promising direction for future work is to explicitly model a distribution over tags for each word type.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The first m pairs have labels yi, whereas for i = m + 1, , n the pairs are unlabeled.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The commentaries in PCC are all of roughly the same length, ranging from 8 to 10 sentences.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The complexity of the quasimonotone search is O(E3 J (R2+LR)).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In the denomi 11 We have two such lists, one containing about 17,000 full names, and another containing frequencies of.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The relevant variables are the set of token-level tags that appear before and after each instance of the ith word type; we denote these context pairs with the set {(tb, ta)} and they are contained in t(\u00e2\u0088\u0092i).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"It is the performance we could achieve if an omniscient observer told us which parser to pick for each of the sentences.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The weak hypothesis can abstain from predicting the label of an instance x by setting h(x) = 0.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We call this technique constituent voting.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Each xt E 2x is the set of features constituting the ith example.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"During the search process, a partial hypothesis is extended by choosing a source sentence position, which has not been aligned with a target sentence position yet.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"If is seen infrequently during training (less than a small count), then will not be selected as a feature and all features in this group are set to 0.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"There has also been work using a bootstrap- ping approach [Brin 98; Agichtein and Gravano 00; Ravichandran and Hovy 02].","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"However, those methods need initial seeds, so the relation between entities has to be known in advance.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"It also incorporates the Good-Turing method (Baayen 1989; Church and Gale 1991) in estimating the likelihoods of previously unseen con\u00c2\u00ad structions, including morphological derivatives and personal names.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"rhetorical analysis We are experimenting with a hybrid statistical and knowledge-based system for discourse parsing and summarization (Stede 2003), (Hanneforth et al. 2003), again targeting the genre of commentaries.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The above auxiliary quantity satisfies the following recursive DP equation: Qe0 (e; C; j) = Initial Skip Verb Final 1.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Section 2.1 describes how BABAR generates training examples to use in the learning process.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The hash variant is a reverse trie with hash tables, a more memory-efficient version of SRILM\u2019s default.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"And time is short.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"I = 1X21 N and N is a &quot;medium&quot; sized number so that it is feasible to collect 0(N) unlabeled examples.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"2.3 Assigning Evidence Values.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This best instance-weighting model beats the equivalant model without instance weights by between 0.6 BLEU and 1.8 BLEU, and beats the log-linear baseline by a large margin.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Statistical methods seem particularly applicable to the problem of unknown-word identification, especially for constructions like names, where the linguistic constraints are minimal, and where one therefore wants to know not only that a particular se\u00c2\u00ad quence of hanzi might be a name, but that it is likely to be a name with some probabil\u00c2\u00ad ity.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"88,962 (spelling,context) pairs were extracted as training data.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The form fmnh, for example, can be understood as the verb \u201clubricated\u201d, the possessed noun \u201cher oil\u201d, the adjective \u201cfat\u201d or the verb \u201cgot fat\u201d.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank, leaving only sections 22 and 23 completely untouched during the development of any of the parsers.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"For example, let us consider a tree set containing trees of the form shown in Figure 4a.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"State will ultimately be used as context in a subsequent query.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We experiment with four values for each hyperparameter resulting in 16 (\u00ce\u00b1, \u00ce\u00b2) combinations: \u00ce\u00b1 \u00ce\u00b2 0.001, 0.01, 0.1, 1.0 0.01, 0.1, 1.0, 10 Iterations In each run, we performed 30 iterations of Gibbs sampling for the type assignment variables W .4 We use the final sample for evaluation.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"8 66.4 52.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Lexical rules are estimated in a similar manner.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The method just described segments dictionary words, but as noted in Section 1, there are several classes of words that should be handled that are not found in a standard dictionary.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The rhetorical structure annotations of PCC have all been converted to URML.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"To make the projection practical, we rely on the twelve universal part-of-speech tags of Petrov et al. (2011).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In addition to the named-entity string (Maury Cooper or Georgia), a contextual predictor was also extracted.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"An IG can be viewed as a CFG in which each nonterminal is associated with a stack.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"What both of these approaches presume is that there is a sin\u00c2\u00ad gle correct segmentation for a sentence, against which an automatic algorithm can be compared.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"However, supervised methods rely on labeled training data, which is time-consuming and expensive to generate.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Here a set is represented by the keyword and the number in parentheses indicates the number of shared NE pair instances.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The computing time is low, since no reordering is carried out.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Queries detect the invalid probability, using the node only if it leads to a longer match.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"On the first of these-the B set-our system had 64% recall and 86% precision; on the second-the C set-it had 33% recall and 19% precision.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Its only purpose is 3 This follows since each \u00ce\u00b8t has St \u00e2\u0088\u0092 1 parameters and.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Similarly, there is no compelling evidence that either of the syllables of f.ifflll binllang2 'betelnut' represents a morpheme, since neither can occur in any context without the other: more likely fjfflll binllang2 is a disyllabic morpheme.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Future extensions of the system might include: 1) An extended translation model, where we use more context to predict a source word.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"A simple extension will be used to handle this problem.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"In both cases, the instanceweighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline, and gains of between 0.6 and 1.8 over an equivalent mixture model (with an identical training procedure but without instance weighting).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Mikheev et al.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"All language model queries issued by machine translation decoders follow a left-to-right pattern, starting with either the begin of sentence token or null context for mid-sentence fragments.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"A better approach would be to distin guish between these cases, possibly by drawing on the vast linguistic work on Arabic connectives (AlBatal, 1990).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Acknowledgments We thank Steven Bethard, Evan Rosen, and Karen Shiells for material contributions to this work.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"This may seem surprising, given the experiments reported in section 4, but the explanation is probably that the non-projective dependencies that can be recovered at all are of the simple kind that only requires a single lift, where the encoding of path information is often redundant.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Although the tag distributions of the foreign words (Eq.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Such sequences are given additional features of A begin, A continue, or A end, and the acronym is given a feature A unique.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Call the crossing constituents A and B.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Backoff-smoothed models estimate this probability based on the observed entry with longest matching history wnf , returning where the probability p(wn|wn\u22121 f ) and backoff penalties b(wn\u22121 i ) are given by an already-estimated model.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We do not adapt the alignment procedure for generating the phrase table from which the TM distributions are derived.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, the characterization given in the main body of the text is correct sufficiently often to be useful.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The counts represent portions of the approximately 44000 constituents hypothesized by the parsers in the development set.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This has the potential drawback of increasing the number of features, which can make MERT less stable (Foster and Kuhn, 2009).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We feel that information from a whole corpus might turn out to be noisy if the documents in the corpus are not of the same genre.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":".","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"annotation guidelines that tell annotators what to do in case of doubt.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Assuming unseen objects within each class are equiprobable, their probabilities are given by the Good-Turing theorem as: cis E( n'J.ls) Po oc N * E(N8ls) (2) where p815 is the probability of one unseen hanzi in class cls, E(n'J.15 ) is the expected number of hanzi in cls seen once, N is the total number of hanzi, and E(N(\/ 5 ) is the expected number of unseen hanzi in class cls.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Since trees in a tree set are adjoined together, the addressing scheme uses a sequence of pairings of the address and name of the elementary tree adjoined at that address.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"6 Joint Segmentation and Parsing.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"There has been a lot of research on such lexical relations, along with the creation of resources such as WordNet.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The availability of comparable corpora is limited, which is a significant limitation on the approach.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"P St = n. \u00ce\u00b2 T VARIABLES \u00cf\u0088 Y W : Word types (W1 ,.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"2.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Methods for expanding the dictionary include, of course, morphological rules, rules for segmenting personal names, as well as numeral sequences, expressions for dates, and so forth (Chen and Liu 1992; Wang, Li, and Chang 1992; Chang and Chen 1993; Nie, Jin, and Hannan 1994).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This significantly underperforms log-linear combination.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"During training, we treat as observed the language word types W as well as the token-level corpus w. We utilize Gibbs sampling to approximate our collapsed model posterior: P (T ,t|W , w, \u00ce\u00b1, \u00ce\u00b2) \u00e2\u0088\u009d P (T , t, W , w|\u00ce\u00b1, \u00ce\u00b2) 0.7 0.6 0.5 0.4 0.3 English Danish Dutch Germany Portuguese Spanish Swedish = P (T , t, W , w, \u00cf\u0088, \u00ce\u00b8, \u00cf\u0086, w|\u00ce\u00b1, \u00ce\u00b2)d\u00cf\u0088d\u00ce\u00b8d\u00cf\u0086 Note that given tag assignments T , there is only one setting of token-level tags t which has mass in the above posterior.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"For simplicity, we assume that OUT is homogeneous.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Considerable amount of work has been done in recent years on the named entity recognition task, partly due to the Message Understanding Conferences (MUC).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Lexicon Feature: The string of the token is used as a feature.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Other kinds of productive word classes, such as company names, abbreviations (termed fijsuolxie3 in Mandarin), and place names can easily be 20 Note that 7 in E 7 is normally pronounced as leO, but as part of a resultative it is liao3..","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The equation for sampling a single type-level assignment Ti is given by, 0.2 0 5 10 15 20 25 30 Iteration Figure 2: Graph of the one-to-one accuracy of our full model (+FEATS) under the best hyperparameter setting by iteration (see Section 5).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"2 70.7 52.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Morphological analyzers for Hebrew that analyze a surface form in isolation have been proposed by Segal (2000), Yona and Wintner (2005), and recently by the knowledge center for processing Hebrew (Itai et al., 2006).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"To define a similarity function between the English and the foreign vertices, we rely on high-confidence word alignments.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"64 76.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Note that since all English vertices were extracted from the parallel text, we will have an initial label distribution for all vertices in Ve.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Instead, we condition on the type-level tag assignments T . Specifically, let St = {i|Ti = t} denote the indices of theword types which have been assigned tag t accord ing to the tag assignments T . Then \u00ce\u00b8t is drawn from DIRICHLET(\u00ce\u00b1, St), a symmetric Dirichlet which only places mass on word types indicated by St. This ensures that each word will only be assigned a single tag at inference time (see Section 4).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This is not to say that a set of standards by which a particular segmentation would count as correct and another incorrect could not be devised; indeed, such standards have been proposed and include the published PRCNSC (1994) and ROCLING (1993), as well as the unpublished Linguistic Data Consortium standards (ca.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"segmentation (Table 2).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"As discussed in more detail in \u00a73, we use two types of vertices in our graph: on the foreign language side vertices correspond to trigram types, while the vertices on the English side are individual word types.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The average agreement among the human judges is .76, and the average agreement between ST and the humans is .75, or about 99% of the interhuman agreement.15 One can better visualize the precision-recall similarity matrix by producing from that matrix a distance matrix, computing a classical metric multidimensional scaling (Torgerson 1958; Becker, Chambers, Wilks 1988) on that dis\u00c2\u00ad tance matrix, and plotting the first two most significant dimensions.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"2.1 Overview.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Evalb, the standard parsing metric, is biased toward such corpora (Sampson and Babarczy, 2003).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"4 To be sure, it is not always true that a hanzi represents a syllable or that it represents a morpheme.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Two common cases are the attribu tive adjective and the process nominal _; maSdar, which can have a verbal reading.4 At tributive adjectives are hard because they are or- thographically identical to nominals; they are inflected for gender, number, case, and definiteness.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"2.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The details are given in (Tillmann, 2000).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Top 5 Bot to m 5 Go ld NN P NN JJ CD NN S RB S PD T # \u00e2\u0080\u009d , 1T W CD W RB NN S VB N NN PR P$ W DT : MD . +P RI OR CD JJ NN S WP $ NN RR B- , $ \u00e2\u0080\u009d . +F EA TS JJ NN S CD NN P UH , PR P$ # . \u00e2\u0080\u009c Table 5: Type-level English POS Tag Ranking: We list the top 5 and bottom 5 POS tags in the lexicon and the predictions of our models under the best hyperparameter setting.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Each word is simply tagged with the semantic classes corresponding to all of its senses.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This highly effective approach is not directly applicable to the multinomial models used for core SMT components, which have no natural method for combining split features, so we rely on an instance-weighting approach (Jiang and Zhai, 2007) to downweight domain-specific examples in OUT.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For a sequence of hanzi that is a possible name, we wish to assign a probability to that sequence qua name.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"7 www.cis.upenn.edu\/\u00e2\u0088\u00bcpdtb\/ 8 www.eml-research.de\/english\/Research\/NLP\/ Downloads had to buy a new car.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For each set, the phrases with bracketed frequencies are considered not paraphrases in the set.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":", December, then the feature MonthName is set to 1.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The features are weighted within a logistic model to give an overall weight that is applied to the phrase pair\u2019s frequency prior to making MAP-smoothed relative-frequency estimates (different weights are learned for each conditioning direction).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"1 55.8 38.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Similarly, hanzi sharing the GHOST radical _m tend to denote spirits and demons, such as _m gui3 'ghost' itself, II: mo2 'demon,' and yan3 'nightmare.'","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Typical data structures are generalized Bloom filters that guarantee a customizable probability of returning the correct answer.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Links can be of two different kinds: anaphoric or bridging (definite noun phrases picking up an antecedent via world-knowledge).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"3.2 Reordering with IBM Style.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"One hybridization strategy is to let the parsers vote on constituents' membership in the hypothesized set.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Keywords with more than one word In the evaluation, we explained that \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d are considered paraphrases.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Such a classification can be seen as a not-always-correct summary of global features.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The same form fmnh can be segmented as f-mnh, f (\u201cthat\u201d) functioning as a reletivizer with the form mnh.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The AdaBoost algorithm was developed for supervised learning.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"68 95.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"We assume that the goal in dependency parsing is to construct a labeled dependency graph of the kind depicted in Figure 1.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The first 3770 trees of the resulting set then were used for training, and the last 418 are used testing.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For corporate suffixes, a list of tokens cslist that occur frequently as the last token of an organization name is collected from the training data.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"When a token fmnh is to be interpreted as the lexeme sequence f\/REL mnh\/VB, the analysis introduces two distinct entities, the relativizer f (\u201cthat\u201d) and the verb mnh (\u201ccounted\u201d), and not as the complex entity \u201cthat counted\u201d.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"This is an iterative method that improves the estimation of the parameters at each iteration.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"La ng ua ge 1T W + P RI O R + F E A T S E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 2 1.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"However, since we want to preserve as much of the original structure as possible, we are interested in finding a transformation that involves a minimal number of lifts.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The German finite verbs 'bin' (second example) and 'konnten' (third example) are too far away from the personal pronouns 'ich' and 'Sie' (6 respectively 5 source sentence positions).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For the PROBING implementation, hash table sizes are in the millions, so the most relevant values are on the right size of the graph, where linear probing wins.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"diesem 3.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We consider the unsupervised POS induction problem without the use of a tagging dictionary.","label":0},{"source_text":"There is no global pruning.","target_text":"During the search process, a partial hypothesis is extended by choosing a source sentence position, which has not been aligned with a target sentence position yet.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Because of this threshold, very few NE instance pairs could be used and hence the variety of phrases was also limited.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The paper is structured as follows.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"See Section 5.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Bikel et al.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We suspect that the higher precision in the disasters domain may be due to its substantially larger training corpus.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"A contextual role represents the role that a noun phrase plays in an event or relationship.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"\u00e2\u0080\u0094ested in correlations between prosody and dis course structure.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Using RandLM and the documented settings (8-bit values and 1 256 false-positive probability), we built a stupid backoff model on the same data as in Section 5.2.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Particular instances of relations are associated with goodness scores.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"However, TRIE partitions storage by n-gram length, so walking the trie reads N disjoint pages.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"These derivation trees show how the composition operations were used to derive the final structures from elementary structures.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In this case, we have no finite-state restrictions for the search space.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Of course, we.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"It was developed in response to the non-terminal\/terminal bias of Evalb, but Clegg and Shepherd (2005) showed that it is also a valuable diagnostic tool for trees with complex deep structures such as those found in the ATB.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"We use MMAX for this annotation as well.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The ability to redistribute belief values across sets rather than individual hypotheses is key.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This is a straightforward technique that is arguably better suited to the adaptation task than the standard method of treating representative IN sentences as queries, then pooling the match results.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"As is standard, we report the greedy one-to-one (Haghighi and Klein, 2006) and the many-to-one token-level accuracy obtained from mapping model states to gold POS tags.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Hence, their relationship to formalisms such as HG's and TAG's is of interest.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"kann 7.nicht 8.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"To improve agreement during the revision process, a dual-blind evaluation was performed in which 10% of the data was annotated by independent teams.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"For simplicity, we assume that OUT is homogeneous.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The restriction can be expressed in terms of the number of uncovered source sentence positions to the left of the rightmost position m in the coverage set.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Two measures that can be used to compare judgments are: 1.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For this example, since the sequence Even News Broadcasting Corp. only appears once in the document, its longest substring that occurs in the same document is News Broadcasting Corp. In this case, News has an additional feature of I begin set to 1, Broadcasting has an additional feature of I continue set to 1, and Corp. has an additional feature of I end set to 1.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We also add an annotation for one-level iDafa (oneLevelIdafa) constructs since they make up more than 75% of the iDafa NPs in the ATB (Gabbard and Kulick, 2008).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"To this end, we picked 100 sentences at random containing 4,372 total hanzi from a test corpus.14 (There were 487 marks of punctuation in the test sentences, including the sentence-final periods, meaning that the average inter-punctuation distance was about 9 hanzi.)","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The orthographic normalization strategy we use is simple.10 In addition to removing all diacritics, we strip instances of taTweel J=J4.i, collapse variants of alif to bare alif,11 and map Ara bic punctuation characters to their Latin equivalents.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The following two error criteria are used in our experiments: mWER: multi-reference WER: We use the Levenshtein distance between the automatic translation and several reference translations as a measure of the translation errors.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"72 78.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The relevance of the distinction between, say, phonological words and, say, dictionary words is shown by an example like rpftl_A :;!:Hfllil zhong1hua2 ren2min2 gong4he2-guo2 (China people republic) 'People's Republic of China.'","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The family name set is restricted: there are a few hundred single-hanzi family names, and about ten double-hanzi ones.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The weak hypothesis can abstain from predicting the label of an instance x by setting h(x) = 0.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"3.1 General Knowledge Sources.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For the disasters domain, 8245 texts were used for training and the 40 test documents contained 447 anaphoric links.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"SSER: subjective sentence error rate: For a more detailed analysis, the translations are judged by a human test person.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"0 X u} \"' o; .2 X X><X X XX X X X X X X x X X X X X x X V X X X X .;t'*- XXX:OX X X X X X X 9 x X X XX XX X X X X X X X XXX:< X X>O<XX>!KXX XI<>< \u00c2\u00bbC X X XX :X: X X \"' X X XX >OO<X>D<XIK X X X X X X --XX\u00c2\u00bb: XXX X X\u00c2\u00bbC X X\u00c2\u00abX...C:XXX X Xll< X X ><XX>IIC:liiC:oiiiiCI--8!X:liiOC!I!S8K X X X 10 100 1000 10000 log(F)_base: R\"2=0.20 (p < 0.005) X 100000 Figure 6 Plot of log frequency of base noun, against log frequency of plural nouns.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We pick the parse that is most similar to the other parses by choosing the one with the highest sum of pairwise similarities.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(5) and ht into Equ.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Its success depends on the two domains being relatively close, and on the OUT corpus not being so large as to overwhelm the contribution of IN.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We can 5 Recall that precision is defined to be the number of correct hits divided by the total number of items.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"We have presented a new method for non-projective dependency parsing, based on a combination of data-driven projective dependency parsing and graph transformation techniques.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"am 11.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Bikel et al.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"In this paper we have proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","label":1},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Sentence (2) and (3) help to disambiguate one way or the other.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We use a patched version of BitPar allowing for direct input of probabilities instead of counts.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Instead of the names of elementary trees of a TAG, the nodes are labeled by a sequence of names of trees in an elementary tree set.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The state in future has not enough work for its many teachers.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Step 4.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"T e r r o r i s m Ca sef ra me Semantic Classes <a ge nt > ass ass ina ted group, human inv esti gat ion int o < N P> event exp lod ed out sid e < N P> building N a t u r a l D i s a s t e r s Ca sef ra me Semantic Classes <a ge nt > inv esti gat ing cau se group, human sur viv or of < N P> event, natphenom hit wit h < N P> attribute, natphenom Figure 3: Semantic Caseframe Expectations Figure 2: Lexical Caseframe Expectations To illustrate how lexical expectations are used, suppose we want to determine whether noun phrase X is the antecedent for noun phrase Y. If they are coreferent, then X and Y should be substitutable for one another in the story.4 Consider these sentences: (S1) Fred was killed by a masked man with a revolver.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This allow the learners to &quot;bootstrap&quot; each other by filling the labels of the instances on which the other side has abstained so far.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The general idea for the knowledge- based part is to have the system use as much information as it can find at its disposal to produce a target representation as specific as possible and as underspecified as necessary.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The weight on each sentence is a value in [0, 1] computed by a perceptron with Boolean features that indicate collection and genre membership.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"In MSA, SVO usually appears in non-matrix clauses.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(Berland and Charniak 99) describe a method for extracting parts of objects from wholes (e.g., &quot;speedometer&quot; from &quot;car&quot;) from a large corpus using hand-crafted patterns.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For derived words that occur in our corpus we can estimate these costs as we would the costs for an underived dictionary entry.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(See also Wu and Fung [1994].)","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We can give a tree pumping lemma for TAG's by adapting the uvwxy-theorem for CFL's since the tree sets of TAG's have independent and context-free paths.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"While the first three models get three to four tags wrong, our best model gets only one word wrong and is the most accurate among the four models for this example.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"7).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Our \u201cProjection\u201d baseline is able to benefit from the bilingual information and greatly improves upon the monolingual baselines, but falls short of the \u201cNo LP\u201d model by 2.5% on an average.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The weight on each sentence is a value in [0, 1] computed by a perceptron with Boolean features that indicate collection and genre membership.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This flexibility, along with the simplicity of implementation and expansion, makes this framework an attractive base for continued research.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Each knowledge source then assigns a probability estimate to each candidate, which represents its belief that the candidate is the antecedent for the anaphor.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"4 69.0 51.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"A straightforward way to find the shortest tour is by trying all possible permutations of the n cities.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In fact, during the first rounds many of the predictions of Th., g2 are zero.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"They considered a semi-supervised POS tagging scenario and showed that one can use a graph over trigram types, and edge weights based on distributional similarity, to improve a supervised conditional random field tagger.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The two new terms force the two classifiers to agree, as much as possible, on the unlabeled examples.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Queries take the form p(wn|wn\u22121 1 ) where wn1 is an n-gram.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"On the other hand, we can expect Head+Path to be the most useful representation for reconstructing the underlying non-projective dependency graph.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The corpus has been annotated with six different types of information, which are characterized in the following subsections.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The edge weights between the foreign language trigrams are computed using a co-occurence based similarity function, designed to indicate how syntactically similar the middle words of the connected trigrams are (\u00a73.2).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"\u00ce\u00b8 has a belief value of 1.0, indicating complete certainty that the correct hypothesis is included in the set, and a plausibility value of 1.0, indicating that there is no evidence for competing hypotheses.5 As evidence is collected and the likely hypotheses are whittled down, belief is redistributed to subsets of \u00ce\u00b8.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The extent to which this constraint is enforced varies greatly across existing methods.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The results in Table 2 were achieved on the development set.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The corpus was wordaligned using both HMM and IBM2 models, and the phrase table was the union of phrases extracted from these separate alignments, with a length limit of 7.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Each decision determines the inclusion or exclusion of a candidate constituent.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We train linear mixture models for conditional phrase pair probabilities over IN and OUT so as to maximize the likelihood of an empirical joint phrase-pair distribution extracted from a development set.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"SRILM\u2019s compact variant, IRSTLM, MITLM, and BerkeleyLM\u2019s sorted variant are all based on this technique.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"An error count of 0:0 is assigned to a perfect translation, and an error count of 1:0 is assigned to a semantically and syntactically wrong translation.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The relevant variables are the set of token-level tags that appear before and after each instance of the ith word type; we denote these context pairs with the set {(tb, ta)} and they are contained in t(\u00e2\u0088\u0092i).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Evaluating parsing results in our joint framework, as argued by Tsarfaty (2006), is not trivial under the joint disambiguation task, as the hypothesized yield need not coincide with the correct one.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The relevant variables are the set of token-level tags that appear before and after each instance of the ith word type; we denote these context pairs with the set {(tb, ta)} and they are contained in t(\u00e2\u0088\u0092i).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"This work was supported in part by the Swedish Research Council (621-2002-4207).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Annotation of syntactic structure for the core corpus has just begun.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In this example there are four \"input characters,\" A, B, C and D, and these map respectively to four \"pronunciations\" a, b, c and d. Furthermore, there are four \"words\" represented in the dictionary.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We have used the Java-based opennlp maximum entropy package1.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"When the maSdar lacks a determiner, the constituent as a whole resem bles the ubiquitous annexation construct \u00ef\u00bf\u00bd ?f iDafa.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Base NPs are the other significant category of nominal phrases.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Te rro ris m Na tur al Dis ast ers mu rde r of < NP > kill ed <p atie nt > <a ge nt > da ma ged wa s inj ure d in < NP > <a ge nt > rep ort ed <a ge nt > add ed <a ge nt > occ urr ed cau se of < NP > <a ge nt > stat ed <a ge nt > add ed <a ge nt > wr eak ed <a ge nt > cro sse d per pet rat ed <p atie nt > con de mn ed <p atie nt > dri ver of < NP > <a ge nt > car ryi ng Figure 1: Caseframe Network Examples Figure 1 shows examples of caseframes that co-occur in resolutions, both in the terrorism and natural disaster domains.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(2) was extended to have an additional, innermost loop over the (3) possible labels.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We used 22 features for the logistic weighting model, divided into two groups: one intended to reflect the degree to which a phrase pair belongs to general language, and one intended to capture similarity to the IN domain.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This decreases the statistical significance of our results compared to those studies.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In the natural disasters domain, agents are often forces of nature, such as hurricanes or wildfires.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"4 65.9 48.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Still, for both human and automatic rhetorical analysis, connectives are the most important source of surface information.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Each distance in the traveling salesman problem now corresponds to the negative logarithm of the product of the translation, alignment and language model probabilities.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Phrase-level granularity distinguishes our work from previous work by Matsoukas et al (2009), who weight sentences according to sub-corpus and genre membership.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We characterize a class of formalisms that have this property in Section 4.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"When a collision occurs, linear probing places the entry to be inserted in the next (higher index) empty bucket, wrapping around as necessary.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"For Arabic we M o d e l S y s t e m L e n g t h L e a f A n c e s t o r Co rpu s Sent Exact E v a l b L P LR F1 T a g % B a s e l i n e 7 0 St an for d (v 1.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"6. 3) all G o l d P O S 7 0 0.7 91 0.825 358 0.7 73 0.818 358 0.8 02 0.836 452 80.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For a language like English, this problem is generally regarded as trivial since words are delimited in English text by whitespace or marks of punctuation.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Finally, the focus\/background partition is annotated, together with the focus question that elicits the corresponding answer.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"pronunciation depends upon word affiliation: tfJ is pronounced deO when it is a prenominal modification marker, but di4 in the word \u00c2\u00a7tfJ mu4di4 'goal'; fl; is normally ganl 'dry,' but qian2 in a person's given name.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"(2010) and the posterior regular- ization HMM of Grac\u00c2\u00b8a et al.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Smith estimates Lotus will make a profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d, our system extracts \u00e2\u0080\u009cSmith esti mates Lotus\u00e2\u0080\u009d as an instance.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The final estimating equation is then: (3) Since the total of all these class estimates was about 10% off from the Turing estimate n1\/N for the probability of all unseen hanzi, we renormalized the estimates so that they would sum to n 1jN.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Collisions between two keys in the table can be identified at model building time.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Entries for 2 < n < N store a vocabulary identifier, probability, backoff, and an index into the n + 1-gram table.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Arabic sentences of up to length 63 would need to be.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The developers suggested some changes, such as building the model from scratch with IRSTLM, but these did not resolve the problem.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"These packages are further described in Section 3.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The algorithm works due to the fact that not all permutations of cities have to be considered explicitly.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Smith estimates Lotus will make a profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d, our system extracts \u00e2\u0080\u009cSmith esti mates Lotus\u00e2\u0080\u009d as an instance.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In the namedentity problem each example is a (spelling,context) pair.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For the translation experiments, Eq. 2 is recursively evaluated.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Particular instances of relations are associated with goodness scores.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The 8 similarity-to-IN features are based on word frequencies and scores from various models trained on the IN corpus: To avoid numerical problems, each feature was normalized by subtracting its mean and dividing by its standard deviation.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The negative logarithm of t0 is reported.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Denote the unthresholded classifiers after t \u2014 1 rounds by git\u20141 and assume that it is the turn for the first classifier to be updated while the second one is kept fixed.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"6 Our knowledge sources return some sort of probability estimate, although in some cases this estimate is not especially well-principled (e.g., the Recency KS).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"When the parser is trained on the transformed data, it will ideally learn not only to construct projective dependency structures but also to assign arc labels that encode information about lifts.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Evaluation of the Segmentation as a Whole.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Two lists, Corporate-Suffix-List (for corporate suffixes) and Person-Prefix-List (for person prefixes), are collected from the training data.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"For this reason, na\u00efve Bayes classifiers are well-matched to this problem.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The first is an evaluation of the system's ability to mimic humans at the task of segmenting text into word-sized units; the second evaluates the proper-name identification; the third measures the performance on morphological analysis.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The SynRole KS computes the relative frequency with which the candidates\u00e2\u0080\u0099 syntactic role (subject, direct object, PP object) appeared in resolutions in the training set.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Moreover, the Stanford parser achieves the most exact Leaf Ancestor matches and tagging accuracy that is only 0.1% below the Bikel model, which uses pre-tagged input.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Previous approaches have tried to find examples that are similar to the target domain.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The number of NE instance pairs used in their experiment is less than half of our method.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This alters generation of T as follows: n P (T |\u00cf\u0088) = n P (Ti|\u00cf\u0088) i=1 Note that this distribution captures the frequency of a tag across word types, as opposed to tokens.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"5.1 Data Sets.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Arabic is a morphologically rich language with a root-and-pattern system similar to other Semitic languages.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"4 65.9 48.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Extract NE instance pairs with contexts First, we extract NE pair instances with their context from the corpus.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The advantage is that we can recombine search hypotheses by dynamic programming.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Unlike Germann et al. (2009), we chose a model size so that all benchmarks fit comfortably in main memory.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"A contextual role represents the role that a noun phrase plays in an event or relationship.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"In the rhetorical tree, nuclearity information is then used to extract a \u00e2\u0080\u009ckernel tree\u00e2\u0080\u009d that supposedly represents the key information from which the summary can be generated (which in turn may involve co-reference information, as we want to avoid dangling pronouns in a summary).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"VBD she added VP PUNC \u00e2\u0080\u009c SBAR IN NP 0 NN.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The corpus-based statistical parsing community has many fast and accurate automated parsing systems, including systems produced by Collins (1997), Charniak (1997) and Ratnaparkhi (1997).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"buy - acquire (5) buy - agree (2) buy - purchase (5) buy - acquisition (7) buy - pay (2)* buy - buyout (3) buy - bid (2) acquire - purchase (2) acquire - acquisition (2) acquire - pay (2)* purchase - acquisition (4) purchase - stake (2)* acquisition - stake (2)* unit - subsidiary (2) unit - parent (5) It is clear that these links form two clusters which are mostly correct.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In our third model GTppp we also add the distinction between general PPs and possessive PPs following Goldberg and Elhadad (2007).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Other good classes include JADE and GOLD; other bad classes are DEATH and RAT.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":":zhong1 : 0.0 tjl :huo2 :0.0 (R:spub:\/ic of Ch:ina) + .,_,...I : jlong4 :0.0 (mUifaty genG181) 0 \u00c2\u00a3: _NC: 40.0 Figure 3 Partial Chinese Lexicon (NC = noun; NP = proper noun).c=- - I \u00e2\u0080\u00a2=- :il: .;ss:;zhangt \u00e2\u0080\u00a2 '-:.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Since we could not bias the subjects towards a particular segmentation and did not presume linguistic sophistication on their part, the instructions were simple: subjects were to mark all places they might plausibly pause if they were reading the text aloud.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The authors provided us with a ratio between TPT and SRI under different conditions. aLossy compression with the same weights. bLossy compression with retuned weights. ditions make the value appropriate for estimating repeated run times, such as in parameter tuning.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The bias of automatic methods in favor of statistical systems seems to be less pronounced on out-of-domain test data.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Again, the idea is that having a picture of syntax, co-reference, and sentence-internal information structure at one\u00e2\u0080\u0099s disposal should aid in finding models of discourse structure that are more explanatory and can be empirically supported.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":".., Tn ) T W \u00cf\u0086 E w : Token word seqs (obs) t : Token tag assigns (det by T ) PARAMETERS \u00cf\u0088 : Lexicon parameters \u00ce\u00b8 : Token word emission parameters \u00cf\u0086 : Token tag transition parameters \u00cf\u0086 \u00cf\u0086 t1 t2 \u00ce\u00b8 \u00ce\u00b8 w1 w2 K \u00cf\u0086 T tm O K \u00ce\u00b8 E wN m N N Figure 1: Graphical depiction of our model and summary of latent variables and parameters.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"f1; ; Jg denotes a coverage set including all positions from the starting position 1 to position J and j 2 fJ \u00f4\u0080\u0080\u0080L; ; Jg.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"A generative model was applied (similar to naive Bayes) with the three labels as hidden vanables on unlabeled examples, and observed variables on (seed) labeled examples.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, there will remain a large number of words that are not readily adduced to any produc\u00c2\u00ad tive pattern and that would simply have to be added to the dictionary.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Supervised learning approaches have advanced the state-of-the-art on a variety of tasks in natural language processing, resulting in highly accurate systems.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We use two common techniques, hash tables and sorted arrays, describing each before the model that uses the technique.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The first point we need to address is what type of linguistic object a hanzi repre\u00c2\u00ad sents.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Table 3 shows BABAR\u00e2\u0080\u0099s performance when the four contextual role knowledge sources are added.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The final estimating equation is then: (3) Since the total of all these class estimates was about 10% off from the Turing estimate n1\/N for the probability of all unseen hanzi, we renormalized the estimates so that they would sum to n 1jN.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The second algorithm extends ideas from boosting algorithms, designed for supervised learning tasks, to the framework suggested by (Blum and Mitchell 98).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"As you can see in the figure, the accuracy for the domain is quite high except for the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set, which contains various expressions representing different relationships for an IE application.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Unfortunately, Yarowsky's method is not well understood from a theoretical viewpoint: we would like to formalize the notion of redundancy in unlabeled data, and set up the learning task as optimization of some appropriate objective function.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"However, there is a crucial difference: the morphological probabilities in their model come from discriminative models based on linear context.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The Stanford parser includes both the manually annotated grammar (\u00c2\u00a74) and an Arabic unknown word model with the following lexical features: 1.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Given the bilingual graph described in the previous section, we can use label propagation to project the English POS labels to the foreign language.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Current state-of-the-art broad-coverage parsers assume a direct correspondence between the lexical items ingrained in the proposed syntactic analyses (the yields of syntactic parse-trees) and the spacedelimited tokens (henceforth, \u2018tokens\u2019) that constitute the unanalyzed surface forms (utterances).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"For example, from the sentence \u00e2\u0080\u009cMr.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Parameter Component As in the standard Bayesian HMM (Goldwater and Griffiths, 2007), all distributions are independently drawn from symmetric Dirichlet distributions: 2 Note that t and w denote tag and word sequences respectively, rather than individual tokens or tags.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In this section, we brie y review our translation approach.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The DL-CoTrain algorithm can be motivated as being a greedy method of satisfying the above 2 constraints.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Many packages perform language model queries.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(2008) reported agreement between the teams (measured with Evalb) at 93.8% F1, the level of the CTB.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The 2nd block contains the IR system, which was tuned by selecting text in multiples of the size of the EMEA training corpus, according to dev set performance.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"As discussed in more detail in \u00a73, we use two types of vertices in our graph: on the foreign language side vertices correspond to trigram types, while the vertices on the English side are individual word types.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Case and Zone of and : Similarly, if (or ) is initCaps, a feature (initCaps, zone) (or (initCaps, zone) ) is set to 1, etc. Token Information: This group consists of 10 features based on the string , as listed in Table 1.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Since guess and gold trees may now have different yields, the question of evaluation is complex.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We pick the parse that is most similar to the other parses by choosing the one with the highest sum of pairwise similarities.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"However their work did not consider other types of lexical expectations (e.g., PP arguments), semantic expectations, or context comparisons like our case- frame network.(Niyu et al., 1998) used unsupervised learning to ac quire gender, number, and animacy information from resolutions produced by a statistical pronoun resolver.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If is one of Monday, Tuesday, . . .","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"On the other hand, we can expect Head+Path to be the most useful representation for reconstructing the underlying non-projective dependency graph.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This modification brings the method closer to the DL-CoTrain algorithm described earlier, and is motivated by the intuition that all three labels should be kept healthily populated in the unlabeled examples, preventing one label from dominating \u2014 this deserves more theoretical investigation.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"A dynamic programming recursion similar to the one in Eq. 2 is evaluated.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Either save money at any cost - or give priority to education.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Statistical methods seem particularly applicable to the problem of unknown-word identification, especially for constructions like names, where the linguistic constraints are minimal, and where one therefore wants to know not only that a particular se\u00c2\u00ad quence of hanzi might be a name, but that it is likely to be a name with some probabil\u00c2\u00ad ity.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"This feature imposes constraints Table 1: Features based on the token string that are based on the probability of each name class during training.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The Stanford parser includes both the manually annotated grammar (\u00c2\u00a74) and an Arabic unknown word model with the following lexical features: 1.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"For each trigram type x2 x3 x4 in a sequence x1 x2 x3 x4 x5, we count how many times that trigram type co-occurs with the different instantiations of each concept, and compute the point-wise mutual information (PMI) between the two.5 The similarity between two trigram types is given by summing over the PMI values over feature instantiations that they have in common.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In the second scheme, Head+Path, we in addition modify the label of every arc along the lifting path from the syntactic to the linear head so that if the original label is p the new label is p\u2193.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Furthermore we expect the label distributions on the foreign to be fairly noisy, because the graph constraints have not been taken into account yet.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Here, all tokeninternal collocations of tags unseen in our training data are pruned away.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In this section, we describe how contextual role knowledge is represented and learned.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We first make use of our morphological analyzer to find all segmentation possibilities by chopping off all prefix sequence possibilities (including the empty prefix) and construct a lattice off of them.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, the company may refer to Company X in one paragraph and Company Y in another.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In the following, we use the notation wi wj to mean that (wi, r, wj) E A; r we also use wi wj to denote an arc with unspecified label and wi \u2014*\u2217 wj for the reflexive and transitive closure of the (unlabeled) arc relation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We have provided methods for handling certain classes of unknown words, and models for other classes could be provided, as we have noted.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Memory-based classifiers for the experiments were created using TiMBL (Daelemans et al., 2003).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In the cotraining case, (Blum and Mitchell 98) argue that the task should be to induce functions Ii and f2 such that So Ii and 12 must (1) correctly classify the labeled examples, and (2) must agree with each other on the unlabeled examples.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"A promising direction for future work is to explicitly model a distribution over tags for each word type.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Note that Zt is a normalization constant that ensures the distribution Dt+i sums to 1; it is a function of the weak hypothesis ht and the weight for that hypothesis at chosen at the tth round.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Performance typically stabilizes across languages after only a few number of iterations.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"7).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The bias of automatic methods in favor of statistical systems seems to be less pronounced on out-of-domain test data.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"9 61.0 44.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Using a treebank grammar, a data-driven lexicon, and a linguistically motivated unknown-tokens handling technique our model outperforms previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","label":1},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The relevance of the distinction between, say, phonological words and, say, dictionary words is shown by an example like rpftl_A :;!:Hfllil zhong1hua2 ren2min2 gong4he2-guo2 (China people republic) 'People's Republic of China.'","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Typical data structures are generalized Bloom filters that guarantee a customizable probability of returning the correct answer.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We briefly review it here for completeness.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The techniques we develop can be extended in a relatively straightforward manner to the more general case when OUT consists of multiple sub-domains.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"KenLM: Faster and Smaller Language Model Queries","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"This limitation is the obstacle to making the technology \u00e2\u0080\u009copen domain\u00e2\u0080\u009d.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Having explained the various layers of annotation in PCC, we now turn to the question what all this might be good for.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"\u00e2\u0080\u009d, because Bush is the first word, the initial caps might be due to its position (as in \u00e2\u0080\u009cThey put a freeze on . . .","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Hyperparameters Our model has two Dirichlet concentration hyperparameters: \u00ce\u00b1 is the shared hyperparameter for the token-level HMM emission and transition distributions.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"hanzi in the various name positions, derived from a million names.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The most frequent NE category pairs are \u00e2\u0080\u009cPerson - Person (209,236), followed by \u00e2\u0080\u009cCountry - Coun- try\u00e2\u0080\u009d (95,123) and \u00e2\u0080\u009cPerson - Country\u00e2\u0080\u009d (75,509).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"While we have minimized forward-looking state in Section 4.1, machine translation systems could also benefit by minimizing backward-looking state.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Apart from MERT difficulties, a conceptual problem with log-linear combination is that it multiplies feature probabilities, essentially forcing different features to agree on high-scoring candidates.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The percentage scores on the axis labels represent the amount of variation in the data explained by the dimension in question.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"IL+-1Proof: Assume a pair of crossing constituents appears in the output of the constituent voting technique using k parsers.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"However, the characterization given in the main body of the text is correct sufficiently often to be useful.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In considering the relationship between formalisms, we show that it is useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees. find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars On the basis of this observation, we describe a class of formalisms which we call Linear Context- Free Rewriting Systems, and show they are recognizable in polynomial time and generate only semilinear languages.","label":1},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For example, we might have VP \u00e2\u0086\u0092 VB NP PP, where the NP is the subject.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We do not experiment with models larger than physical memory in this paper because TPT is unreleased, factors such as disk speed are hard to replicate, and in such situations we recommend switching to a more compact representation, such as RandLM.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"While processing the source sentence monotonically, the initial state I is entered whenever there are no uncovered positions to the left of the rightmost covered position.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"then define the best segmentation to be the cheapest or best path in Id(I) o D* (i.e., Id(I) composed with the transitive closure of 0).6 Consider the abstract example illustrated in Figure 2.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Each feature concept is akin to a random variable and its occurrence in the text corresponds to a particular instantiation of that random variable.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"On the other hand, in a translation system one probably wants to treat this string as a single dictionary word since it has a conventional and somewhat unpredictable translation into English.","label":0},{"source_text":"Here we present two algorithms.","target_text":"AdaBoost finds a weighted combination of simple (weak) classifiers, where the weights are chosen to minimize a function that bounds the classification error on a set of training examples.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"), which precludes a single universal approach to adaptation.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Keyword detection error Even if a keyword consists of a single word, there are words which are not desirable as keywords for a domain.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Since the parameter and token components will remain fixed throughout experiments, we briefly describe each.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"markBaseNP indicates these non-recursive nominal phrases.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"A different notion of information structure, is used in work such as that of (?), who tried to characterize felicitous constituent ordering (theme choice, in particular) that leads to texts presenting information in a natural, \u00e2\u0080\u009cflowing\u00e2\u0080\u009d way rather than with abrupt shifts of attention.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The terms on the right-hand-side denote the type-level and token-level probability terms respectively.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"0 68.4 48.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"However, it is robust, efficient, and easy to implement.4 To perform the maximization in (7), we used the popular L-BFGS algorithm (Liu and Nocedal, 1989), which requires gradient information.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"from the subset of the United Informatics corpus not used in the training of the models.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Third, all remaining anaphora are evaluated by 11 different knowledge sources: the four contextual role knowledge sources just described and seven general knowledge sources.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"When OUT is large and distinct, its contribution can be controlled by training separate IN and OUT models, and weighting their combination.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"an event.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The CoBoost algorithm just described is for the case where there are two labels: for the named entity task there are three labels, and in general it will be useful to generalize the CoBoost algorithm to the multiclass case.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The government has to make a decision, and do it quickly.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Even this may be nondeterministic, in case the graph contains several non-projective arcs whose lifts interact, but we use the following algorithm to construct a minimal projective transformation D0 = (W, A0) of a (nonprojective) dependency graph D = (W, A): The function SMALLEST-NONP-ARC returns the non-projective arc with the shortest distance from head to dependent (breaking ties from left to right).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"On one end of the spectrum are clustering approaches that assign a single POS tag to each word type (Schutze, 1995; Lamar et al., 2010).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As indicated in Figure 1(c), apart from this correct analysis, there is also the analysis taking B ri4 as a word (e.g., a common abbreviation for Japan), along with X:\u00c2\u00a5 wen2zhangl 'essay\/ and f!!.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The complexity of the quasimonotone search is O(E3 J (R2+LR)).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"While size of the resulting transducers may seem daunting-the segmenter described here, as it is used in the Bell Labs Mandarin TTS system has about 32,000 states and 209,000 arcs-recent work on minimization of weighted machines and transducers (cf.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"attaching to terms denoting human beings.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, for our purposes it is not sufficient to repre\u00c2\u00ad sent the morphological decomposition of, say, plural nouns: we also need an estimate of the cost of the resulting word.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For \u00c3\u0086 = 0, no new target word is generated, while an additional source sentence position is covered.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Unlike GCFG's, however, the composition operations of LCFRS's are restricted to be linear (do not duplicate unboundedly large structures) and nonerasing (do not erase unbounded structures, a restriction made in most modern transformational grammars).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"In an experiment on automatic rhetorical parsing, the RST-annotations and PoS tags were used by (Reitter 2003) as a training corpus for statistical classification with Support Vector Machines.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"A direct-mapped cache makes BerkeleyLM faster on repeated queries, but their fastest (scrolling) cached version is still slower than uncached PROBING, even on cache-friendly queries.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In this paper k = 3 (the three labels are person, organization, location), and we set a = 0.1.","label":0},{"source_text":"There is no global pruning.","target_text":"The goal of machine translation is the translation of a text given in some source language into a target language.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"No: Predecessor coverage set Successor coverage set 1 (f1; ;mg n flg ; l0) !","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Since our goal is to perform well under these measures we will similarly treat constituents as the minimal substructures for combination.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The core of Yarowsky's algorithm is as follows: where h is defined by the formula in equation 2, with counts restricted to training data examples that have been labeled in step 2.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"21 84.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We train linear mixture models for conditional phrase pair probabilities over IN and OUT so as to maximize the likelihood of an empirical joint phrase-pair distribution extracted from a development set.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Except our own and MENE + reference resolution, the results in Table 6 are all official MUC7 results.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"6One of our experimental settings lacks document boundaries, and we used this approximation in both settings for consistency.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We define the following function: If Zco is small, then it follows that the two classifiers must have a low error rate on the labeled examples, and that they also must give the same label on a large number of unlabeled instances.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Interpolation search formalizes the notion that one opens a dictionary near the end to find the word \u201czebra.\u201d Initially, the algorithm knows the array begins at b +\u2014 0 and ends at e +\u2014 |A|\u22121.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Interestingly, Chang et al. report 80.67% recall and 91.87% precision on an 11,000 word corpus: seemingly, our system finds as many names as their system, but with four times as many false hits.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The weak learner for two-class problems computes a weak hypothesis h from the input space into the reals (h : 2x -4 R), where the sign4 of h(x) is interpreted as the predicted label and the magnitude I h(x)I is the confidence in the prediction: large numbers for I h(x)I indicate high confidence in the prediction, and numbers close to zero indicate low confidence.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Obviously, the presence of a title after a potential name N increases the probability that N is in fact a name.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"State is implemented in their scrolling variant, which is a trie annotated with forward and backward pointers.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Parameter Component As in the standard Bayesian HMM (Goldwater and Griffiths, 2007), all distributions are independently drawn from symmetric Dirichlet distributions: 2 Note that t and w denote tag and word sequences respectively, rather than individual tokens or tags.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"However, the next step is clearly different.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The corresponding token words w are drawn conditioned on t and \u00ce\u00b8.2 Our full generative model is given by: K P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) = n (P (\u00cf\u0086t|\u00ce\u00b1)P (\u00ce\u00b8t|T , \u00ce\u00b1)) t=1 The transition distribution \u00cf\u0086t for each tag t is drawn according to DIRICHLET(\u00ce\u00b1, K ), where \u00ce\u00b1 is the shared transition and emission distribution hyperparameter.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"For each trigram type x2 x3 x4 in a sequence x1 x2 x3 x4 x5, we count how many times that trigram type co-occurs with the different instantiations of each concept, and compute the point-wise mutual information (PMI) between the two.5 The similarity between two trigram types is given by summing over the PMI values over feature instantiations that they have in common.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"With the additional assumption that (s, t) can be restricted to the support of co(s, t), this is equivalent to a \u201cflat\u201d alternative to (6) in which each non-zero co(s, t) is set to one.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The portions of information in the large window can be individually clicked visible or invisible; here we have chosen to see (from top to bottom) \u00e2\u0080\u00a2 the full text, \u00e2\u0080\u00a2 the annotation values for the activated annotation set (co-reference), \u00e2\u0080\u00a2 the actual annotation tiers, and \u00e2\u0080\u00a2 the portion of text currently \u00e2\u0080\u0098in focus\u00e2\u0080\u0099 (which also appears underlined in the full text).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For fair comparison, we have tabulated all results with the size of training data used (Table 5 and Table 6).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Email: rlls@bell-labs.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Unigram records store probability, backoff, and an index in the bigram table.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"On the other hand, only very restricted reorderings are necessary, e.g. for the translation direction from Table 2: Coverage set hypothesis extensions for the IBM reordering.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"0 55.3 34.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Named Entity Recognition: A Maximum Entropy Approach Using Global Information","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"First, we identify sources of syntactic ambiguity understudied in the existing parsing literature.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"I \u00e2\u0080\u00a2 JAPANS :rl4 .\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\)\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"o\u00c2\u00b7'\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\:J\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7 \u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7'\\; . '.:: ..........0 6.51 9.51 : jj \/ JAPANESE OCTOPUS 10\u00c2\u00b728i\u00c2\u00a3 :_nc HOW SAY f B :rl4 :il: :wen2 t '- \u00e2\u0080\u00a2 :zhang!","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Since the early days of statistical NLP, researchers have observed that a part-of-speech tag distribution exhibits \u00e2\u0080\u009cone tag per discourse\u00e2\u0080\u009d sparsity \u00e2\u0080\u0094 words are likely to select a single predominant tag in a corpus, even when several tags are possible.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"In our grammar, features are realized as annotations to basic category labels.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Our coreference resolver also incorporates an existential noun phrase recognizer and a DempsterShafer probabilistic model to make resolution decisions.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"\u00e2\u0080\u00a2 Anaphoric links: the annotator is asked to specify whether the anaphor is a repetition, partial repetition, pronoun, epithet (e.g., Andy Warhol \u00e2\u0080\u0093 the PopArt artist), or is-a (e.g., Andy Warhol was often hunted by photographers.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We further report SYNCS, the parsing metric of Cohen and Smith (2007), to facilitate the comparison.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"more frequently than is done in English.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"computing the precision of the other's judgments relative to this standard.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The availability of comparable corpora is limited, which is a significant limitation on the approach.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"(1997; 1999) do have some statistics that show how IdentiFinder performs when the training data is reduced.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, if a token starts with a capital letter and ends with a period (such as Mr.), then the feature InitCapPeriod is set to 1, etc. First Word: This feature group contains only one feature firstword.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We tokenize MWUs and their POS tags; this reduces the tag set size to 12.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"4.5 Transliterations of Foreign Words.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In contrast to these approaches, our method directly incorporates these constraints into the structure of the model.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Here, \u00e2\u0080\u009cEG\u00e2\u0080\u009d represents \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Email: cls@bell-labs.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Unexpectedly, because the ministries of treasury and education both had prepared the teacher plan together.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Still, for both human and automatic rhetorical analysis, connectives are the most important source of surface information.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"4 65.9 48.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For example, hanzi containing the INSECT radical !R tend to denote insects and other crawling animals; examples include tr wal 'frog,' feng1 'wasp,' and !Itt she2 'snake.'","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Apart from MERT difficulties, a conceptual problem with log-linear combination is that it multiplies feature probabilities, essentially forcing different features to agree on high-scoring candidates.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"In the cases where isolated constituent precision is larger than 0.5 the affected portion of the hypotheses is negligible.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"With the same Corporate- Suffix-List and Person-Prefix-List used in local features, for a token seen elsewhere in the same document with one of these suffixes (or prefixes), another feature Other-CS (or Other-PP) is set to 1.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Therefore, a populated probing hash table consists of an array of buckets that contain either one entry or are empty.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Table 1 shows four words \u00e2\u0080\u009c 0 Indeed NN Indeed Saddamwhose unvocalized surface forms 0 an are indistinguishable.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"4 Traditional Arabic linguistic theory treats both of these types as subcategories of noun \u00ef\u00bf\u00bd '.i . Figure 1: The Stanford parser (Klein and Manning, 2002) is unable to recover the verbal reading of the unvocalized surface form 0 an (Table 1).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"3An English sentence with ambiguous PoS assignment can be trivially represented as a lattice similar to our own, where every pair of consecutive nodes correspond to a word, and every possible PoS assignment for this word is a connecting arc.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Here, the pruning threshold t0 = 10:0 is used.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Table 3: Training and test conditions for the Verbmobil task (*number of words without punctuation marks).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Unigram records store probability, backoff, and an index in the bigram table.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Annotation of syntactic structure for the core corpus has just begun.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Unfortunately, modifying the model to account for these kind of dependencies is not at all straightforward.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Recent work has made significant progress on unsupervised POS tagging (Me\u00c2\u00b4rialdo, 1994; Smith and Eisner, 2005; Haghighi and Klein, 2006; Johnson,2007; Goldwater and Griffiths, 2007; Gao and John son, 2008; Ravi and Knight, 2009).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"This overview is illustrated in Figure 1.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This is akin to PoS tags sequences induced by different parses in the setup familiar from English and explored in e.g.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"phrase (markContainsVerb).","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Blum and Mitchell 98) describe learning in the following situation: X = X1 X X2 where X1 and X2 correspond to two different &quot;views&quot; of an example.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In this paper, we have presented a new, e\u00c3\u0086cient DP-based search procedure for statistical machine translation.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"3.1 Maximum Entropy.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"4 53.7 43.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(Hearst 92) describes a method for extracting hyponyms from a corpus (pairs of words in &quot;isa&quot; relations).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"A graphical depiction of our model as well as a summary of random variables and parameters can be found in Figure 1.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Language models that contain wi must also contain prefixes wi for 1 G i G k. Therefore, when the model is queried for p(wnjwn\u22121 1 ) but the longest matching suffix is wnf , it may return state s(wn1) = wnf since no longer context will be found.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"First, we evaluated BABAR using only the seven general knowledge sources.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"All commentaries have been tagged with part-of-speech information using Brants\u00e2\u0080\u0099 TnT1 tagger and the Stuttgart\/Tu\u00c2\u00a8bingen Tag Set automatic analysis was responsible for this decision.)","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This variant is tested in the experiments below.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"It is natural that the larger the data in the domain, the more keywords are found.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Depending on the threshold t0, the search algorithm may miss the globally optimal path which typically results in additional translation errors.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"It then computes a normalized Levenshtein edit distance between the extracted chain and the reference.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"With the exception of the Dutch data set, no other processing is performed on the annotated tags.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The paper is structured as follows.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, this result is consistent with the results of ex\u00c2\u00ad periments discussed in Wu and Fung (1994).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Different sentence structure and rich target language morphology are two reasons for this.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Thus in an English sentence such as I'm going to show up at the ACL one would reasonably conjecture that there are eight words separated by seven spaces.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"For this reason, na\u00efve Bayes classifiers are well-matched to this problem.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"2 62.6 45.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Thus it is possible, for illustration, to look for a noun phrase (syntax tier) marked as topic (information structure tier) that is in a bridging relation (co-reference tier) to some other noun phrase.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"While we cannot prove there are no such useful features on which one should condition trust, we can give some insight into why the features we explored offered no gain.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Ex: The regime gives itself the right...","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"For instance, the sentence Similar improvements in haemoglobin levels were reported in the scientific literature for other epoetins would likely be considered domain-specific despite the presence of general phrases like were reported in.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We contrasted formalisms such as CFG's, HG's, TAG's and MCTAG's, with formalisms such as IG's and unificational systems such as LFG's and FUG's.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Though we are not able to calculate their memory usage on our model, results reported in their paper suggest lower memory consumption than TRIE on large-scale models, at the expense of CPU time.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In sequential tagging models such as (Adler and Elhadad, 2006; Bar-Haim et al., 2007; Smith et al., 2005) weights are assigned according to a language model The input for the joint task is a sequence W = w1, ... , wn of space-delimited tokens.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For illustration, an English translation of one of the commentaries is given in Figure 1.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"While we cannot prove there are no such useful features on which one should condition trust, we can give some insight into why the features we explored offered no gain.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"9 www.ling.unipotsdam.de\/sfb\/ Figure 2: Screenshot of Annis Linguistic Database 3.3 Symbolic and knowledge-based.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For example, in Information Retrieval (IR), we have to match a user\u00e2\u0080\u0099s query to the expressions in the desired documents, while in Question Answering (QA), we have to find the answer to the user\u00e2\u0080\u0099s question even if the formulation of the answer in the document is different from the question.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Nonetheless, parse quality is much lower in the joint model because a lattice is effectively a long sentence.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"83 77.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"com t 600 Mountain Avenue, 2c278, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"7.96 5.55 1 l...................................................................................................................................................................................................J..","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We believe that the underlying principles of the maximum entropy framework are suitable for exploiting information from diverse sources.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Recent work by Finkel and Manning (2009) which re-casts Daum\u00b4e\u2019s approach in a hierarchical MAP framework may be applicable to this problem.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"As discussed in Section 3, independent paths in tree sets, rather than the path complexity, may be crucial in characterizing semilinearity and polynomial time recognition.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"A different knowledge source, called CFSemCFSem, compares the semantic expectations of the caseframe that extracts the anaphor with the semantic expectations of the caseframe that extracts the candidate.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"While the paper mentioned a sorted variant, code was never released.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Hearst 92) describes a method for extracting hyponyms from a corpus (pairs of words in &quot;isa&quot; relations).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"For example, one parser could be more accurate at predicting noun phrases than the other parsers.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Although these authors report better gains than ours, they are with respect to a non-adapted baseline.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We use a squared loss to penalize neighboring vertices that have different label distributions: kqi \u2212 qjk2 = Ey(qi(y) \u2212 qj(y))2, and additionally regularize the label distributions towards the uniform distribution U over all possible labels Y.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Here, $ is the sentence boundary symbol, which is thought to be at position 0 in the target sentence.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Similarly, there is no compelling evidence that either of the syllables of f.ifflll binllang2 'betelnut' represents a morpheme, since neither can occur in any context without the other: more likely fjfflll binllang2 is a disyllabic morpheme.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"2.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The attachment in such cases encompasses a long distance dependency that cannot be captured by Markovian processes that are typically used for morphological disambiguation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"A minimal requirement for building a Chinese word segmenter is obviously a dictionary; furthermore, as has been argued persuasively by Fung and Wu (1994), one will perform much better at segmenting text by using a dictionary constructed with text of the same genre as the text to be segmented.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Other work includes transferring latent topic distributions from source to target language for LM adaptation, (Tam et al., 2007) and adapting features at the sentence level to different categories of sentence (Finch and Sumita, 2008).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Let s = a + b.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Step 2.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"handled given appropriate models.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The baseline alignment model does not permit that a source word is aligned to two or more target words, e.g. for the translation direction from German toEnglish, the German compound noun 'Zahnarztter min' causes problems, because it must be translated by the two target words dentist's appointment.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Parent Head Modif er Dir # gold F1 Label # gold F1 NP NP TAG R 946 0.54 ADJP 1216 59.45 S S S R 708 0.57 SBAR 2918 69.81 NP NP ADJ P R 803 0.64 FRAG 254 72.87 NP NP N P R 2907 0.66 VP 5507 78.83 NP NP SBA R R 1035 0.67 S 6579 78.91 NP NP P P R 2713 0.67 PP 7516 80.93 VP TAG P P R 3230 0.80 NP 34025 84.95 NP NP TAG L 805 0.85 ADVP 1093 90.64 VP TAG SBA R R 772 0.86 WHN P 787 96.00 S VP N P L 961 0.87 (a) Major phrasal categories (b) Major POS categories (c) Ten lowest scoring (Collins, 2003)-style dependencies occurring more than 700 times Table 8: Per category performance of the Berkeley parser on sentence lengths \u00e2\u0089\u00a4 70 (dev set, gold segmentation).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Mohri [1995]) shows promise for improving this situation.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Because conjunctions are elevated in the parse trees when they separate recursive constituents, we choose the right sister instead of the category of the next word.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Members of LCFRS whose operations have this property can be translated into the ILFP notation (Rounds, 1985).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In (b) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the kidnapping victims, but in (c) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the armed men.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Our monolingual similarity function (for connecting pairs of foreign trigram types) is the same as the one used by Subramanya et al. (2010).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Since guess and gold trees may now have different yields, the question of evaluation is complex.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"With each iteration more examples are assigned labels by both classifiers, while a high level of agreement (> 94%) is maintained between them.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Two general approaches are presented and two combination techniques are described for each approach.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Two annotators received training with the RST definitions and started the process with a first set of 10 texts, the results of which were intensively discussed and revised.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Note that in line 4 the last visited position for the successor hypothesis must be m. Otherwise , there will be four uncovered positions for the predecessor hypothesis violating the restriction.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"A contextual rule considers words surrounding the string in the sentence in which it appears (e.g., a rule that any proper name modified by an appositive whose head is president is a person).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In many cases these failures in recall would be fixed by having better estimates of the actual prob\u00c2\u00ad abilities of single-hanzi words, since our estimates are often inflated.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A TAG consists of a finite set of elementary trees that are either initial trees or auxiliary trees.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Other strategies could readily 6 As a reviewer has pointed out, it should be made clear that the function for computing the best path is. an instance of the Viterbi algorithm.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Adding the isolated constituents to our hypothesis parse could increase our expected recall, but in the cases we investigated it would invariably hurt our precision more than we would gain on recall.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Given parameter estimates, the label for a test example x is defined as We should note that the model in equation 9 is deficient, in that it assigns greater than zero probability to some feature combinations that are impossible.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(3)In sentence (1), McCann can be a person or an orga nization.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"98 15.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"While it is possible to derive a closed form solution for this convex objective function, it would require the inversion of a matrix of order |Vf|.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The equation for sampling a single type-level assignment Ti is given by, 0.2 0 5 10 15 20 25 30 Iteration Figure 2: Graph of the one-to-one accuracy of our full model (+FEATS) under the best hyperparameter setting by iteration (see Section 5).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For E(ni1s), then, we substitute a smooth S against the number of class elements.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"In MUC6, the best result is achieved by SRA (Krupka, 1995).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"A possible probabilistic model for assigning probabilities to complex analyses of a surface form may be and indeed recent sequential disambiguation models for Hebrew (Adler and Elhadad, 2006) and Arabic (Smith et al., 2005) present similar models.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"83 77.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The number of paths that can be dependent is bounded by the grammar (in fact the maximum cardinality of a tree set determines this bound).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"This can be seen as a rough approximation of Yarowsky and Ngai (2001).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Jiang and Zhai (2007) suggest the following derivation, making use of the true OUT distribution po(s, t): where each fi(s, t) is a feature intended to charac- !0\u02c6 = argmax pf(s, t) log p\u03b8(s|t) (8) terize the usefulness of (s, t), weighted by Ai. \u03b8 s,t pf(s, t)po(s, t) log p\u03b8(s|t) The mixing parameters and feature weights (col- != argmax po (s, t) lectively 0) are optimized simultaneously using dev- \u03b8 s,t pf(s, t)co(s, t) log p\u03b8(s|t), set maximum likelihood as before: !\ufffdargmax po (s, t) ! \u03b8 s,t \ufffd\u02c6 = argmax \u02dcp(s, t) log p(s|t; 0).","label":0},{"source_text":"Here we present two algorithms.","target_text":"(If fewer than n rules have Precision greater than pin, we 3Note that taking tlie top n most frequent rules already makes the method robut to low count events, hence we do not use smoothing, allowing low-count high-precision features to be chosen on later iterations. keep only those rules which exceed the precision threshold.) pm,n was fixed at 0.95 in all experiments in this paper.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This solution also obviates the need to perform word sense disambiguation.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Keyword detection error Even if a keyword consists of a single word, there are words which are not desirable as keywords for a domain.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This is an issue that we have not addressed at the current stage of our research.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Thus our proposed model is a proper model assigning probability mass to all (7r, L) pairs, where 7r is a parse tree and L is the one and only lattice that a sequence of characters (and spaces) W over our alpha-beth gives rise to.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In this section, we brie y review our translation approach.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We have studied the structural descriptions (tree sets) that can be assigned by various grammatical systems, and classified these formalisms on the basis of two features: path complexity; and path independence.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"95 B a s e li n e ( S e lf t a g ) 70 a l l B i k e l ( v 1 . 2 ) B a s e l i n e ( P r e t a g ) 7 0 a l l G o l d P O S 70 0.7 70 0.801 278 0.7 52 0.794 278 0.7 71 0.804 295 0.7 52 0.796 295 0.7 75 0.808 309 77.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In our implementation, we make perhaps the simplest choice of weak hypothesis.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"5.2 Discussion.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"When the connective is an adverbial, there is much less clarity as to the range of the spans.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In the appositive case, the contextual predictor was the head of the modifying appositive (president in the Maury Cooper example); in the second case, the contextual predictor was the preposition together with the noun it modifies (plant_in in the Georgia example).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"These packages are further described in Section 3.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"In this paper we study the problem of using a parallel corpus from a background domain (OUT) to improve performance on a target domain (IN) for which a smaller amount of parallel training material\u2014though adequate for reasonable performance\u2014is also available.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Hence, for each token, one of the four features zone-TXT, zone- HL, zone-DATELINE, or zone-DD is set to 1, and the other 3 are set to 0.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"It is likely that the more complex cases, where path information could make a difference, are beyond the reach of the parser in most cases.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Backoff-smoothed models estimate this probability based on the observed entry with longest matching history wnf , returning where the probability p(wn|wn\u22121 f ) and backoff penalties b(wn\u22121 i ) are given by an already-estimated model.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We again adopt an approach where we alternate between two classifiers: one classifier is modified while the other remains fixed.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"3.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Once HMM parameters (\u00ce\u00b8, \u00cf\u0086) are drawn, a token-level tag and word sequence, (t, w), is generated in the standard HMM fashion: a tag sequence t is generated from \u00cf\u0086.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Table 3 shows BABAR\u00e2\u0080\u0099s performance when the four contextual role knowledge sources are added.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"With a human evaluation we also showed that ATB inter-annotator agreement remains low relative to the WSJ corpus.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The estimation of the probabilities in the model is carried out as shown in Equation 4.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"3.2 Reordering with IBM Style.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"JA DE G O L D G R AS S SI C K NE SS DE AT H R A T 14.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"If these sets do not overlap, then the words cannot be coreferent.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We extend the Stanford parser to accept pre-generated lattices, where each word is represented as a finite state automaton.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"There have been other kinds of efforts to discover paraphrase automatically from corpora.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We can model this probability straightforwardly enough with a probabilistic version of the grammar just given, which would assign probabilities to the individual rules.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"While we have minimized forward-looking state in Section 4.1, machine translation systems could also benefit by minimizing backward-looking state.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(2) was extended to have an additional, innermost loop over the (3) possible labels.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"At the phrasal level, we remove all function tags and traces.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Compared to decoding, this task is cache-unfriendly in that repeated queries happen only as they naturally occur in text.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The second algorithm builds on a boosting algorithm called AdaBoost (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"needs to be in initCaps to be considered for this feature.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"It is possible one could produce better models by introducing features describing constituents and their contexts because one parser could be much better than the majority of the others in particular situations.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The most informative scheme, Head+Path, gives the highest scores, although with respect to Head the difference is not statistically significant, while the least informative scheme, Path \u2013 with almost the same performance on treebank transformation \u2013 is significantly lower (p < 0.01).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Fig.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The transition from f, to a final state transduces c to the grammatical tag PL with cost cost(unseen(f,)): cost(i\u00c2\u00a5JJ1l.ir,) == cost(i\u00c2\u00a5JJ1l.)","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Simply assigning to each word its most frequent associated tag in a corpus achieves 94.6% accuracy on the WSJ portion of the Penn Treebank.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses two methods to identify anaphors that can be easily and reliably resolved with their antecedent: lexical seeding and syntactic seeding.","label":0},{"source_text":"There is no global pruning.","target_text":"A straightforward way to find the shortest tour is by trying all possible permutations of the n cities.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"na me =>2 ha nzi fa mi ly 2 ha nzi gi ve n 5.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Co-occurrences among the particles themselves are subject to further syntactic and lexical constraints relative to the stem.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"However, our full model takes advantage of word features not present in Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The most informative scheme, Head+Path, gives the highest scores, although with respect to Head the difference is not statistically significant, while the least informative scheme, Path \u2013 with almost the same performance on treebank transformation \u2013 is significantly lower (p < 0.01).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The reason why we did not train with both MUC6 and MUC7 training data at the same time is because the task specifications for the two tasks are not identical.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"It would have therefore also been possible to use the integer programming (IP) based approach of Ravi and Knight (2009) instead of the feature-HMM for POS induction on the foreign side.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The most frequent NE category pairs are \u00e2\u0080\u009cPerson - Person (209,236), followed by \u00e2\u0080\u009cCountry - Coun- try\u00e2\u0080\u009d (95,123) and \u00e2\u0080\u009cPerson - Country\u00e2\u0080\u009d (75,509).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"4 53.7 43.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"For one thing, it is not clear who is to receive settlements or what should happen in case not enough teachers accept the offer of early retirement.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"So, who won the competition?","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"0 68.4 48.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"To eliminate such sequences, we define a transition probability between word classes to be equal to 1 if the sequence is admissible, and 0 otherwise.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"It is difficult to motivate these days why one ministry should be exempt from cutbacks \u00e2\u0080\u0094 at the expense of the others.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For even larger models, we recommend RandLM; the memory consumption of the cache is not expected to grow with model size, and it has been reported to scale well.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"A receives a votes, and B receives b votes.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The correct ambiguity resolution of the syntactic level therefore helps to resolve the morphological one, and vice versa.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"It also uses less memory, with 8 bytes of overhead per entry (we store 16-byte entries with m = 1.5); linked list implementations hash set and unordered require at least 8 bytes per entry for pointers.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The cities of the traveling salesman problem correspond to source Table 1: DP algorithm for statistical machine translation.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"gaolxing4 'happy' => F.i'JF.i'J Jl!","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This model is easily incorporated into the segmenter by building a WFST restrict\u00c2\u00ad ing the names to the four licit types, with costs on the arcs for any particular name summing to an estimate of the cost of that name.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"49 99.","label":0},{"source_text":"There is no global pruning.","target_text":"Skipped (K): The translation of up to one word may be postponed . Verb (V): The translation of up to two words may be anticipated.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Memory usage is the same as with binary search and lower than with set.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The NE tagger is a rule-based system with 140 NE categories [Sekine et al. 2004].","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"This group contains a large number of features (one for each token string present in the training data).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"However, this argument is only plausible if the formal framework allows non-projective dependency structures, i.e. structures where a head and its dependents may correspond to a discontinuous constituent.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Then we ran binary search to determine the least amount of memory with which it would run.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Up to now, most IE researchers have been creating paraphrase knowledge (or IE patterns) by hand and for specific tasks.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"\u201cin the shadow\u201d.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We define the following function: If Zco is small, then it follows that the two classifiers must have a low error rate on the labeled examples, and that they also must give the same label on a large number of unlabeled instances.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Ex: The government said it ...","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The natural language processing community is in the strong position of having many available approaches to solving some of its most fundamental problems.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For the purposes of EM, the &quot;observed&quot; data is {(xi, Ya\u2022 \u2022 \u2022 (xrn, Yrn), xfil, and the hidden data is {ym+i y}.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"1 61.7 37.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Keyword detection error Even if a keyword consists of a single word, there are words which are not desirable as keywords for a domain.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(4) gives In order to minimize Zt, at each iteration the final algorithm should choose the weak hypothesis (i.e., a feature xt) which has values for W+ and W_ that minimize Equ.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Finally, we would like to note that it is possible to devise similar algorithms based with other objective functions than the one given in Equ.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Figure 3 Figure 1.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Consider for a set of constituents the isolated constituent precision parser metric, the portion of isolated constituents that are correctly hypothesized.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Figure 1: Translation of PCC sample commentary (STTS)2.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Taking only the highest frequency rules is much &quot;safer&quot;, as they tend to be very accurate.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Note that the good classes JADE, GOLD and GRASS have lower costs than the bad classes SICKNESS, DEATH and RAT, as desired, so the trend observed for the results of this method is in the right direction.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"1 55.8 38.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Note that Wang, Li, and Chang's.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"But Arabic contains a variety of linguistic phenomena unseen in English.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"27 80.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The Levenshtein distance between the automatic translation and each of the reference translations is computed, and the minimum Levenshtein distance is taken.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Thus at each iteration the method induces at most n x k rules, where k is the number of possible labels (k = 3 in the experiments in this paper). step 3.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Our work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In recent years, coreference resolvers have been evaluated as part of MUC6 and MUC7 (MUC7 Proceedings, 1998).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This Good\u00c2\u00ad Turing estimate of p(unseen(f,) If,) can then be used in the normal way to define the probability of finding a novel instance of a construction in ir, in a text: p(unseen(f,)) = p(unseen(f,) I f,) p(fn Here p(ir,) is just the probability of any construction in ft as estimated from the frequency of such constructions in the corpus.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The kind of annotation work presented here would clearly benefit from the emergence of standard formats and tag sets, which could lead to sharable resources of larger size.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Therefore, we only score guess\/gold pairs with identical character yields, a condition that allows us to measure parsing, tagging, and segmentation accuracy by ignoring whitespace.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(In this figure eps is c) be implemented, though, such as a maximal-grouping strategy (as suggested by one reviewer of this paper); or a pairwise-grouping strategy, whereby long sequences of unattached hanzi are grouped into two-hanzi words (which may have some prosodic motivation).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"All three models evaluated in this paper incorrectly analyze the constituent as iDafa; none of the models attach the attributive adjectives properly.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The first term in the objective function is the graph smoothness regularizer which encourages the distributions of similar vertices (large wij) to be similar.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"May 1995).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"It is possible one could produce better models by introducing features describing constituents and their contexts because one parser could be much better than the majority of the others in particular situations.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Of course, we.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The average agreement among the human judges is .76, and the average agreement between ST and the humans is .75, or about 99% of the interhuman agreement.15 One can better visualize the precision-recall similarity matrix by producing from that matrix a distance matrix, computing a classical metric multidimensional scaling (Torgerson 1958; Becker, Chambers, Wilks 1988) on that dis\u00c2\u00ad tance matrix, and plotting the first two most significant dimensions.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"So, it is too costly to make IE technology \u00e2\u0080\u009copen- domain\u00e2\u0080\u009d or \u00e2\u0080\u009con-demand\u00e2\u0080\u009d like IR or QA.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The features are weighted within a logistic model to give an overall weight that is applied to the phrase pair\u2019s frequency prior to making MAP-smoothed relative-frequency estimates (different weights are learned for each conditioning direction).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Semilinearity and the closely related constant growth property (a consequence of semilinearity) have been discussed in the context of grammars for natural languages by Joshi (1983\/85) and Berwick and Weinberg (1984).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This FSA I can be segmented into words by composing Id(I) with D*, to form the WFST shown in Figure 2(c), then selecting the best path through this WFST to produce the WFST in Figure 2(d).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"As Figure 1 shows, this word has no high-confidence alignment in the Italian-English bitext.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The authors acknowledge the support of the NSF (CAREER grant IIS0448168, and grant IIS 0904684).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"This scenario is applicable to a large set of languages and has been considered by a number of authors in the past (Alshawi et al., 2000; Xi and Hwa, 2005; Ganchev et al., 2009).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We define the lattice L to be the concatenation of the lattices Li corresponding to the input words wi (s.t.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"4.3 Translation Experiments.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The fraction of buckets that are empty is m\u22121 m , so average lookup time is O( m 1) and, crucially, constant in the number of entries.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"TPT has theoretically better locality because it stores ngrams near their suffixes, thereby placing reads for a single query in the same or adjacent pages.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The particular classifier used depends upon the noun.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In the following, we assume that this word joining has been carried out.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"They contain about 200M words (25M, 110M, 40M and 19M words, respectively).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We have presented a method for unsupervised part- of-speech tagging that considers a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In MUC6, the best result is achieved by SRA (Krupka, 1995).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The local feature groups are: Non-Contextual Feature: This feature is set to 1 for all tokens.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"All three models evaluated in this paper incorrectly analyze the constituent as iDafa; none of the models attach the attributive adjectives properly.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"11 www.ling.unipotsdam.de\/sfb\/projekt a3.php 12 This step was carried out in the course of the diploma thesis work of David Reitter (2003), which de serves special mention here.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"3.2 Results.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Ends with the feminine affix :: p. 4.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We call these N \u2212 1 words state.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"On the other hand, we can expect Head+Path to be the most useful representation for reconstructing the underlying non-projective dependency graph.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"In the first stage, we run a single step of label propagation, which transfers the label distributions from the English vertices to the connected foreign language vertices (say, Vf\ufffd) at the periphery of the graph.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"It's not clear how to apply these methods in the unsupervised case, as they required cross-validation techniques: for this reason we use the simpler smoothing method shown here. input to the unsupervised algorithm is an initial, &quot;seed&quot; set of rules.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In the following, we assume that this word joining has been carried out.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Linear probing hash tables must have more buckets than entries, or else an empty bucket will never be found.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"One striking example is the error reduction for Spanish, which reduces error by 36.5% and 24.7% for the best and median settings respectively.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"This is especially true in the case of quotations\u00e2\u0080\u0094which are common in the ATB\u00e2\u0080\u0094where (1) will follow a verb like (2) (Figure 1).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"From this we see that a finer-grained model for parser combination, at least for the features we have examined, will not give us any additional power.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Such ambiguities cause discrepancies between token boundaries (indexed as white spaces) and constituent boundaries (imposed by syntactic categories) with respect to a surface form.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"At most one feature in this group will be set to 1.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"67 95.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"To make the projection practical, we rely on the twelve universal part-of-speech tags of Petrov et al. (2011).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The tool we use is MMAX8, which has been specifically designed for marking co-reference.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Keyword detection error Even if a keyword consists of a single word, there are words which are not desirable as keywords for a domain.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The highestorder N-gram array omits backoff and the index, since these are not applicable.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Models To assess the marginal utility of each component of the model (see Section 3), we incremen- tally increase its sophistication.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"= p(fj je) max \u00c3\u0086;e00 j02Cnfjg np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j 0 )o: The DP equation is evaluated recursively for each hypothesis (e0; e; C; j).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The negative logarithm of t0 is reported.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Our work is motivated by the observation that contextual roles can be critically important in determining the referent of a noun phrase.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"na me =>1 ha nzi fa mi ly 1 ha nzi gi ve n 4.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In the ATB, :: b asta\u00e2\u0080\u0099adah is tagged 48 times as a noun and 9 times as verbal noun.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The final strong hypothesis, denoted 1(x), is then the sign of a weighted sum of the weak hypotheses, 1(x) = sign (Vii atht(x)), where the weights at are determined during the run of the algorithm, as we describe below.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Gan's solution depends upon a fairly sophisticated language model that attempts to find valid syntactic, semantic, and lexical relations between objects of various linguistic types (hanzi, words, phrases).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The second row represents the performance of the median hyperparameter setting.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"No: Predecessor coverage set Successor coverage set 1 (f1; ;mg n flg ; l0) !","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We then show that the combining techniques presented above give better parsing accuracy than any of the individual parsers.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We considered using the MUC6 and MUC7 data sets, but their training sets were far too small to learn reliable co-occurrence statistics for a large set of contextual role relationships.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Thus our proposed model is a proper model assigning probability mass to all (7r, L) pairs, where 7r is a parse tree and L is the one and only lattice that a sequence of characters (and spaces) W over our alpha-beth gives rise to.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Certainly these linguistic factors increase the difficulty of syntactic disambiguation.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"rhetorical analysis We are experimenting with a hybrid statistical and knowledge-based system for discourse parsing and summarization (Stede 2003), (Hanneforth et al. 2003), again targeting the genre of commentaries.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"splitPUNC restores the convention of the WSJ.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The authors acknowledge the support of the NSF (CAREER grant IIS0448168, and grant IIS 0904684).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This section measures performance on shared tasks in order of increasing complexity: sparse lookups, evaluating perplexity of a large file, and translation with Moses.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"In application settings, this may be a profitable strategy.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The algorithm builds two classifiers iteratively: each iteration involves minimization of a continuously differential function which bounds the number of examples on which the two classifiers disagree.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Moses sets the cache size parameter to 50 so we did as well; the resulting cache size is 2.82 GB.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In considering the relationship between formalisms, we show that it is useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees. find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars On the basis of this observation, we describe a class of formalisms which we call Linear Context- Free Rewriting Systems, and show they are recognizable in polynomial time and generate only semilinear languages.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"They are set to fixed constants.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Preprocessing the raw trees improves parsing performance considerably.9 We first discard all trees dominated by X, which indicates errors and non-linguistic text.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Learned Tag Prior (PRIOR) We next assume there exists a single prior distribution \u00cf\u0088 over tag assignments drawn from DIRICHLET(\u00ce\u00b2, K ).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Step 3.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Thus, we feel fairly confident that for the examples we have considered from Gan's study a solution can be incorporated, or at least approximated, within a finite-state framework.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"There are two key benefits of this model architecture.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"58 95.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"To show that the derivation tree set of a TAG is a local set, nodes are labeled by pairs consisting of the name of an elementary tree and the address at which it was adjoined, instead of labelling edges with addresses.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Finally, we provide a realistic eval uation in which segmentation is performed both in a pipeline and jointly with parsing (\u00c2\u00a76).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"(3)In sentence (1), McCann can be a person or an orga nization.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Second, we treat the projected labels as features in an unsupervised model (\u00a75), rather than using them directly for supervised training.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In section 4 we evaluate these transformations with respect to projectivized dependency treebanks, and in section 5 they are applied to parser output.","label":0},{"source_text":"There is no global pruning.","target_text":"6.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For each extension a new position is added to the coverage set.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Each xii is a member of X, where X is a set of possible features.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"5 67.3 55.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"3 68.4 49.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"6 Conclusions.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The general idea for the knowledge- based part is to have the system use as much information as it can find at its disposal to produce a target representation as specific as possible and as underspecified as necessary.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"\u00e2\u0080\u009d).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"When SRILM estimates a model, it sometimes removes n-grams but not n + 1-grams that extend it to the left.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"lla\/llb and 14a\/14b respectively).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"72 78.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Despite their simplicity, uni- gram weights have been shown as an effective feature in segmentation models (Dyer, 2009).13 The joint parser\/segmenter is compared to a pipeline that uses MADA (v3.0), a state-of-the-art Arabic segmenter, configured to replicate ATB segmentation (Habash and Rambow, 2005).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"6 Results and Analysis.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Assuming unseen objects within each class are equiprobable, their probabilities are given by the Good-Turing theorem as: cis E( n'J.ls) Po oc N * E(N8ls) (2) where p815 is the probability of one unseen hanzi in class cls, E(n'J.15 ) is the expected number of hanzi in cls seen once, N is the total number of hanzi, and E(N(\/ 5 ) is the expected number of unseen hanzi in class cls.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"For the experiments, we use a simple preprocessing step.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"For robustness reasons, the parser may output a set of dependency trees instead of a single tree. most dependent of the next input token, dependency type features are limited to tokens on the stack.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The same numbers were used for each data structure.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Besides the lack of a clear definition of what constitutes a correct segmentation for a given Chinese sentence, there is the more general issue that the test corpora used in these evaluations differ from system to system, so meaningful comparison between systems is rendered even more difficult.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For example, syntactic decoders (Koehn et al., 2007; Dyer et al., 2010; Li et al., 2009) perform dynamic programming parametrized by both backward- and forward-looking state.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"This result suggests that all of contextual role KSs can provide useful information for resolving anaphora.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"31 75.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"There has been recent interest in the application of Indexed Grammars (IG's) to natural languages.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"genitiveMark indicates recursive NPs with a indefinite nominal left daughter and an NP right daughter.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"An important aspect of the DempsterShafer model is that it operates on sets of hypotheses.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The feature-HMM model works better for all languages, generalizing the results achieved for English by Berg-Kirkpatrick et al. (2010).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Considerable amount of work has been done in recent years on the named entity recognition task, partly due to the Message Understanding Conferences (MUC).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"We will briefly discuss this point in Section 3.1.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Therefore, we only score guess\/gold pairs with identical character yields, a condition that allows us to measure parsing, tagging, and segmentation accuracy by ignoring whitespace.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"There are clearly eight orthographic words in the example given, but if one were doing syntactic analysis one would probably want to consider I'm to consist of two syntactic words, namely I and am.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Note that while the standard HMM, has O(K n) emission parameters, our model has O(n) effective parameters.3 Token Component Once HMM parameters (\u00cf\u0086, \u00ce\u00b8) have been drawn, the HMM generates a token-level corpus w in the standard way: P (w, t|\u00cf\u0086, \u00ce\u00b8) = P (T , W , \u00ce\u00b8, \u00cf\u0088, \u00cf\u0086, t, w|\u00ce\u00b1, \u00ce\u00b2) = P (T , W , \u00cf\u0088|\u00ce\u00b2) [Lexicon] \u00ef\u00a3\u00ab n n \u00ef\u00a3\u00ad (w,t)\u00e2\u0088\u0088(w,t) j \u00ef\u00a3\u00b6 P (tj |\u00cf\u0086tj\u00e2\u0088\u00921 )P (wj |tj , \u00ce\u00b8tj )\u00ef\u00a3\u00b8 P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) [Parameter] P (w, t|\u00cf\u0086, \u00ce\u00b8) [Token] We refer to the components on the right hand side as the lexicon, parameter, and token component respectively.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"More examples are shown in Figure 5.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For each trigram type x2 x3 x4 in a sequence x1 x2 x3 x4 x5, we count how many times that trigram type co-occurs with the different instantiations of each concept, and compute the point-wise mutual information (PMI) between the two.5 The similarity between two trigram types is given by summing over the PMI values over feature instantiations that they have in common.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"By establishing significantly higher parsing baselines, we have shown that Arabic parsing performance is not as poor as previously thought, but remains much lower than English.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Note that it is in precision that our over\u00c2\u00ad all performance would appear to be poorer than the reported performance of Chang et al., yet based on their published examples, our system appears to be doing better precisionwise.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"\u00e2\u0080\u00a2 Anaphoric links: the annotator is asked to specify whether the anaphor is a repetition, partial repetition, pronoun, epithet (e.g., Andy Warhol \u00e2\u0080\u0093 the PopArt artist), or is-a (e.g., Andy Warhol was often hunted by photographers.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For each domain, phrases which contain the same keyword are gathered to build a set of phrases (Step 3).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"0 X u} \"' o; .2 X X><X X XX X X X X X X x X X X X X x X V X X X X .;t'*- XXX:OX X X X X X X 9 x X X XX XX X X X X X X X XXX:< X X>O<XX>!KXX XI<>< \u00c2\u00bbC X X XX :X: X X \"' X X XX >OO<X>D<XIK X X X X X X --XX\u00c2\u00bb: XXX X X\u00c2\u00bbC X X\u00c2\u00abX...C:XXX X Xll< X X ><XX>IIC:liiC:oiiiiCI--8!X:liiOC!I!S8K X X X 10 100 1000 10000 log(F)_base: R\"2=0.20 (p < 0.005) X 100000 Figure 6 Plot of log frequency of base noun, against log frequency of plural nouns.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"(7) \u03c6 s,t This is a somewhat less direct objective than used by Matsoukas et al, who make an iterative approximation to expected TER.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"7).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"4 70.4 46.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The Hebrew token \u2018bcl\u20191, for example, stands for the complete prepositional phrase 'We adopt here the transliteration of (Sima\u2019an et al., 2001).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"each word in the lexicon whether or not each string is actually an instance of the word in question.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Different sentence structure and rich target language morphology are two reasons for this.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Matsoukas et al (2009) generalize it by learning weights on sentence pairs that are used when estimating relative-frequency phrase-pair probabilities.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The present proposal falls into the last group.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Entries landing in the same bucket are said to collide.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Their results are then compared with the results of an automatic segmenter.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We use graph-based label propagation for cross-lingual knowledge transfer and use the projected labels as features in an unsupervised model (Berg- Kirkpatrick et al., 2010).","label":0},{"source_text":"There is no global pruning.","target_text":"E.g. when 'Zahnarzttermin' is aligned to dentist's, the extended lexicon model might learn that 'Zahnarzttermin' actuallyhas to be aligned to both dentist's and ap pointment.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"A Hebrew surface token may have several readings, each of which corresponding to a sequence of segments and their corresponding PoS tags.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"SRILM\u2019s compact variant has an incredibly expensive destructor, dwarfing the time it takes to perform translation, and so we also modified Moses to avoiding the destructor by calling exit instead of returning normally.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"0 57.3 51.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We use graph-based label propagation for cross-lingual knowledge transfer and use the projected labels as features in an unsupervised model (Berg- Kirkpatrick et al., 2010).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This larger corpus was kindly provided to us by United Informatics Inc., R.O.C. a set of initial estimates of the word frequencies.9 In this re-estimation procedure only the entries in the base dictionary were used: in other words, derived words not in the base dictionary and personal and foreign names were not used.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Experiments are presented in section 4.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We follow the guidelines developed in the TIGER project (Brants et al. 2002) for syntactic annotation of German newspaper text, using the Annotate3 tool for interactive construction of tree structures.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Alon Lavie advised on this work.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"While we have minimized forward-looking state in Section 4.1, machine translation systems could also benefit by minimizing backward-looking state.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In the remainder of the paper, we outline how a class of Linear Context-Free Rewriting Systems (LCFRS's) may be defined and sketch how semilinearity and polynomial recognition of these systems follows.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Our implementation permits jumping to any n-gram of any length with a single lookup; this appears to be unique among language model implementations.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Evaluation of links A link between two sets is considered correct if the majority of phrases in both sets have the same meaning, i.e. if the link indicates paraphrase.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"6. 3) all G o l d P O S 7 0 0.7 91 0.825 358 0.7 73 0.818 358 0.8 02 0.836 452 80.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Space- or punctuation-delimited * 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Segments with the same surface form but different PoS tags are treated as different lexemes, and are represented as separate arcs (e.g. the two arcs labeled neim from node 6 to 7).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"(2009) on Portuguese (Grac\u00c2\u00b8a et al.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"0 . 8 3 1 0.859 496 76.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The larger sets are more accurate than the small sets.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Data We use the Hebrew Treebank, (Sima\u2019an et al., 2001), provided by the knowledge center for processing Hebrew, in which sentences from the daily newspaper \u201cHa\u2019aretz\u201d are morphologically segmented and syntactically annotated.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"example, in Northern Mandarin dialects there is a morpheme -r that attaches mostly to nouns, and which is phonologically incorporated into the syllable to which it attaches: thus men2+r (door+R) 'door' is realized as mer2.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"He used an additional maximum entropy classifier that tries to correct mistakes by using reference resolution.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"This aspect of the formalism is both linguistically and computationally important.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"MENE without Proteus, however, did not do very well and only achieved an F measure of 84.22% (Borthwick, 1999).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The dev corpus was taken from the NIST05 evaluation set, augmented with some randomly-selected material reserved from the training set.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Table 1 briefly describes the seven syntactic heuristics used by BABAR to resolve noun phrases.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Specifically, the lexicon is generated as: P (T , W |\u00cf\u0088) =P (T )P (W |T ) Word Type Features (FEATS): Past unsupervised POS work have derived benefits from features on word types, such as suffix and capitalization features (Hasan and Ng, 2009; Berg-Kirkpatrick et al.,2010).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The relevant variables are the set of token-level tags that appear before and after each instance of the ith word type; we denote these context pairs with the set {(tb, ta)} and they are contained in t(\u00e2\u0088\u0092i).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We generate these caseframes automatically by running AutoSlog over the training corpus exhaustively so that it literally generates a pattern to extract every noun phrase in the corpus.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Translation errors are reported in terms of multireference word error rate (mWER) and subjective sentence error rate (SSER).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"To do this, the x's and y's are stored in the next 2ni + 2n2 tapes, and M goes to a universal state.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"These operations, as we see below, are restricted to be size preserving (as in the case of concatenation in CFG) which implies that they will be linear and non-erasing.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Papers that use this method or minor variants thereof include Liang (1986), Li et al.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The parser switching oracle is the upper bound on the accuracy that can be achieved on this set in the parser switching framework.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"We believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The search starts in the hypothesis (I; f;g; 0).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Ex: Mr. Bush disclosed the policy by reading it...","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"N, portion of examples on which both classifiers give a label rather than abstaining), and the proportion of these examples on which the two classifiers agree.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Current state-of-the-art broad-coverage parsers assume a direct correspondence between the lexical items ingrained in the proposed syntactic analyses (the yields of syntactic parse-trees) and the spacedelimited tokens (henceforth, \u2018tokens\u2019) that constitute the unanalyzed surface forms (utterances).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"For our experiments we also report the mean of precision and recall, which we denote by (P + R)I2 and F-measure.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Because we don\u2019t have a separate development set, we used the training set to select among them and found 0.2 to work slightly better than 0.1 and 0.3.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We train and test on the CoNLL-X training set.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The annotator can then \u00e2\u0080\u009cclick away\u00e2\u0080\u009d those words that are here not used as connectives (such as the conjunction und (\u00e2\u0080\u0098and\u00e2\u0080\u0099) used in lists, or many adverbials that are ambiguous between connective and discourse particle).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"However, we needed to restrict ourselves to these languages in order to be able to evaluate the performance of our approach.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For example, the passive voice pattern \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d and the active voice pattern \u00e2\u0080\u009ckidnapped <direct object>\u00e2\u0080\u009d are merged into a single normalized pattern \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d.2 For the sake of sim plicity, we will refer to these normalized extraction patterns as caseframes.3 These caseframes can capture two types of contextual role information: (1) thematic roles corresponding to events (e.g, \u00e2\u0080\u009c<agent> kidnapped\u00e2\u0080\u009d or \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d), and (2) predicate-argument relations associated with both verbs and nouns (e.g., \u00e2\u0080\u009ckidnapped for <np>\u00e2\u0080\u009d or \u00e2\u0080\u009cvehicle with <np>\u00e2\u0080\u009d).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In the second part of the experiment, we applied the inverse transformation based on breadth-first search under the three different encoding schemes.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The first setting uses the European Medicines Agency (EMEA) corpus (Tiedemann, 2009) as IN, and the Europarl (EP) corpus (www.statmt.org\/europarl) as OUT, for English\/French translation in both directions.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"However, lazy mapping is generally slow because queries against uncached pages must wait for the disk.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"RandLM is the clear winner in RAM utilization, but is also slower and lower quality.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"First, a non-anaphoric NP classifier identifies definite noun phrases that are existential, using both syntactic rules and our learned existential NP recognizer (Bean and Riloff, 1999), and removes them from the resolution process.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The state in future has not enough work for its many teachers.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"More complex approaches such as the relaxation technique have been applied to this problem Fan and Tsai (1988}.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Crucially, the conventional orthographic form of MSA text is unvocalized, a property that results in a deficient graphical representation.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"0 57.3 51.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The sentences in the corpus were tagged by a transformation-based chunker and an NE tagger.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"a classifier.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Conditioned on T , features of word types W are drawn.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The particular classifier used depends upon the noun.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"For example, the formalisms in the hierarchy described above generate semilinear languages although their path sets become increasingly more complex as one moves up the hierarchy.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Restrictions: Quasi-monotone Search The above search space is still too large to allow the translation of a medium length input sentence.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"This paper does not necessarily reflect the position of the U.S. Government.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Gabbard and Kulick (2008) show that there is significant attachment ambiguity associated with iDafa, which occurs in 84.3% of the trees in our development set.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Quantization can be improved by jointly encoding probability and backoff.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"By applying an inverse transformation to the output of the parser, arcs with non-standard labels can be lowered to their proper place in the dependency graph, giving rise 1The dependency graph has been modified to make the final period a dependent of the main verb instead of being a dependent of a special root node for the sentence. to non-projective structures.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"J:j:l :zhongl :0.0 ;m,Jlong4 :0.0 (mHHaryg9tltHBI) \u00c2\u00a3: _ADV: 5.98 \u00c2\u00a5 :hua2:o.o E :_NC: 4.41 :mln2:o.o mm : guo2 : 0.0 (RopubllcofChlna) .....,.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The results of this experiment can be seen in Table 5.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Fort= 1,...,T:","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We insisted that each structure dominates a bounded number of (not necessarily adjacent) substrings.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"All the links in the \u00e2\u0080\u009cCC-domain are shown in Step 4 in subsection 3.2.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"termined by the category of the word that follows it.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We checked whether the discovered links are listed in WordNet.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The linear LM (lin lm), TM (lin tm) and MAP TM (map tm) used with non-adapted counterparts perform in all cases slightly worse than the log-linear combination, which adapts both LM and TM components.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Given around 90,000 unlabeled examples, the methods described in this paper classify names with over 91% accuracy.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The translation of one position in the source sentence may be postponed for up to L = 3 source positions, and the translation of up to two source positions may be anticipated for at most R = 10 source positions.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The approach uses both spelling and contextual rules.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The price shocked her.), or same-kind (e.g., Her health insurance paid for the hospital fees, but the automobile insurance did not cover the repair.).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"During coreference resolution, BABAR checks (1) whether the anaphor is among the lexical expectations for the caseframe that extracts the candidate antecedent, and (2) whether the candidate is among the lexical expectations for the caseframe that extracts the anaphor.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But we follow the more direct adaptation of Evalb suggested by Tsarfaty (2006), who viewed exact segmentation as the ultimate goal.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"In addition to using the simple features directly, we also trained an SVM classifier with these features to distinguish between IN and OUT phrase pairs.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In such cases we assign all of the estimated probability mass to the form with the most likely pronunciation (determined by inspection), and assign a very small probability (a very high cost, arbitrarily chosen to be 40) to all other variants.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"For unaligned words, we set the tag to the most frequent tag in the corresponding treebank.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"An alternate approximation to (8) would be to let w,\\(s, t) directly approximate p\u02c6I(s, t).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For example, ... fraud related to work on a federally funded sewage plant in Georgia In this case, Georgia is extracted: the NP containing it is a complement to the preposition in; the PP headed by in modifies the NP a federally funded sewage plant, whose head is the singular noun plant.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The accuracy results for segmentation, tagging and parsing using our different models and our standard data split are summarized in Table 1.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"For other languages, we use the CoNLL-X multilingual dependency parsing shared task corpora (Buchholz and Marsi, 2006) which include gold POS tags (used for evaluation).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The aforementioned surface form bcl, for example, may also stand for the lexical item \u201conion\u201d, a Noun.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The terrorism examples reflect fairly obvious relationships: people who are murdered are killed; agents that \u00e2\u0080\u009creport\u00e2\u0080\u009d things also \u00e2\u0080\u009cadd\u00e2\u0080\u009d and \u00e2\u0080\u009cstate\u00e2\u0080\u009d things; crimes that are \u00e2\u0080\u009cperpetrated\u00e2\u0080\u009d are often later \u00e2\u0080\u009ccondemned\u00e2\u0080\u009d.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"A detailed description of the search procedure used is given in this patent.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"A more forceful approach for encoding sparsity is posterior regularization, which constrains the posterior to have a small number of expected tag assignments (Grac\u00c2\u00b8a et al., 2009).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The Verbmobil task is an appointment scheduling task.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"First, in section 4, we evaluate the graph transformation techniques in themselves, with data from the Prague Dependency Treebank and the Danish Dependency Treebank.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"5 67.3 55.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Another way to interpret this is that less than 5% of the correct constituents are missing from the hypotheses generated by the union of the three parsers.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The system will then look for sequences of initial capitalized words that match the acronyms found in the whole document.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"An inverted alignment is defined as follows: inverted alignment: i ! j = bi: Target positions i are mapped to source positions bi.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Thus, the effects of spontaneous speech are present in the corpus, e.g. the syntactic structure of the sentence is rather less restricted, however the effect of speech recognition errors is not covered.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Previous work that involves the gathering of information from the whole document often uses a secondary classifier, which corrects the mistakes of a primary sentence- based classifier.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"This assumption, however, is not inherent to type-based tagging models.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"When translating the sentence monotonically from left to right, the translation of the German finite verb 'kann', which is the left verbal brace in this case, is postponed until the German noun phrase 'mein Kollege' is translated, which is the subject of the sentence.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Evaluation Metrics We report three metrics to evaluate tagging performance.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Table 2 shows BABAR\u00e2\u0080\u0099s performance.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Skipped (K): The translation of up to one word may be postponed . Verb (V): The translation of up to two words may be anticipated.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"All the links in the \u00e2\u0080\u009cCC-domain are shown in Step 4 in subsection 3.2.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Figure 3 shows a small fragment of the WFST encoding the dictionary, containing both entries forjust discussed, g:t\u00c2\u00a5 zhonglhua2 min2guo2 (China Republic) 'Republic of China,' and i\u00c2\u00a5inl.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"These conaUses lossy compression. bThe 8-bit quantized variant returned incorrect probabilities as explained in Section 3.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In addition, there are several approaches to non-projective dependency parsing that are still to be evaluated in the large (Covington, 1990; Kahane et al., 1998; Duchier and Debusmann, 2001; Holan et al., 2001; Hellwig, 2003).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"mein 5.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(4) gives In order to minimize Zt, at each iteration the final algorithm should choose the weak hypothesis (i.e., a feature xt) which has values for W+ and W_ that minimize Equ.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In order to pass these constraints onto the parser, the lexical rules in the grammar are of the form pi \u2014* (si, pi) Parameter Estimation The grammar probabilities are estimated from the corpus using simple relative frequency estimates.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"rently, some annotations (in particular the connectives and scopes) have already moved beyond the core corpus; the others will grow step by step.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For verbs we add two features.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We can make several observations on the cause of errors.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"These packages are further described in Section 3.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"8 66.4 52.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"By contrast, when we turn to a comparison of the three encoding schemes it is hard to find any significant differences, and the overall impression is that it makes little or no difference which encoding scheme is used, as long as there is some indication of which words are assigned their linear head instead of their syntactic head by the projective parser.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Evaluation Metrics We report three metrics to evaluate tagging performance.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Time includes all queries but excludes random number generation and data structure population.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"From the discussion so far it is clear that a number of formalisms involve some type of context-free rewriting (they have derivation trees that are local sets).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This is not to say that a set of standards by which a particular segmentation would count as correct and another incorrect could not be devised; indeed, such standards have been proposed and include the published PRCNSC (1994) and ROCLING (1993), as well as the unpublished Linguistic Data Consortium standards (ca.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The derivation trees of a MCTAG are similar to those of a TAG.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Yet, some hanzi are far more probable in women's names than they are in men's names, and there is a similar list of male-oriented hanzi: mixing hanzi from these two lists is generally less likely than would be predicted by the independence model.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Our oracles took advantage of the labeled treebanks: While we tried to minimize the number of free parameters in our model, there are a few hyperparameters that need to be set.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The development of the very first Hebrew Treebank (Sima\u2019an et al., 2001) called for the exploration of general statistical parsing methods, but the application was at first limited.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The core of Yarowsky's algorithm is as follows: where h is defined by the formula in equation 2, with counts restricted to training data examples that have been labeled in step 2.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Our coreference resolver also incorporates an existential noun phrase recognizer and a DempsterShafer probabilistic model to make resolution decisions.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Some of these approaches (e.g., Lin, Chiang, and Su [1993]) attempt to identify unknown words, but do not ac\u00c2\u00ad tually tag the words as belonging to one or another class of expression.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Label propagation is used to propagate these tags inwards and results in tag distributions for the middle word of each Italian trigram.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Once we figure out the important word (e.g. keyword), we believe we can capture the meaning of the phrase by the keyword.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"mark- ContainsVerb is especially effective for distinguishing root S nodes of equational sentences.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In the next section we present an alternative approach that builds two classifiers while attempting to satisfy the above constraints as much as possible.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We show the results of three of the experiments we conducted to measure isolated constituent precision under various partitioning schemes.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Each learner is free to pick the labels for these instances.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We propose a limit of 70 words for Arabic parsing evaluations.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In such cases we use the non-pruned lattice including all (possibly ungrammatical) segmentation, and let the statistics (including OOV) decide.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Specifically, the lexicon is generated as: P (T , W |\u00cf\u0088) =P (T )P (W |T ) Word Type Features (FEATS): Past unsupervised POS work have derived benefits from features on word types, such as suffix and capitalization features (Hasan and Ng, 2009; Berg-Kirkpatrick et al.,2010).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The intuition here is that the role of a discourse marker can usually be de 9 Both the corpus split and pre-processing code are avail-.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"(1) CEO of McCann . . .","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"With the exception of the Dutch data set, no other processing is performed on the annotated tags.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The idea is to have a pipeline of shallow-analysis modules (tagging, chunk- ing, discourse parsing based on connectives) and map the resulting underspecified rhetorical tree (see Section 2.4) into a knowledge base that may contain domain and world knowledge for enriching the representation, e.g., to resolve references that cannot be handled by shallow methods, or to hypothesize coherence relations.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We see from these results that the behavior of the parametric techniques are robust in the presence of a poor parser.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"When extracting the vector t, used to compute the constraint feature from the graph, we tried three threshold values for r (see Eq.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"(2010)\u00e2\u0080\u0099s richest model: optimized via either EM or LBFGS, as their relative performance depends on the language.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We received submissions from 14 groups from 11 institutions, as listed in Figure 2.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"There has been recent interest in the application of Indexed Grammars (IG's) to natural languages.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Instead, we focused on phrases and set the frequency threshold to 2, and so were able to utilize a lot of phrases while minimizing noise.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In speech recognition the arcs of the lattice are typically weighted in order to indicate the probability of specific transitions.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"While building a machine translation system is a serious undertaking, in future we hope to attract more newcomers to the field by keeping the barrier of entry as low as possible.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"(1999) did not report using any dictionaries, but mentioned in a footnote that they have added list membership features, which have helped marginally in certain domains.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"There is no global pruning.","target_text":"These alignment models are similar to the concept of hidden Markov models (HMM) in speech recognition.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Although matching is done at the sentence level, this information is subsequently discarded when all matches are pooled.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The details are given in (Tillmann, 2000).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"All the NE pair instances which co-occur separated by at most 4 chunks are collected along with information about their NE types and the phrase between the NEs (the \u00e2\u0080\u0098context\u00e2\u0080\u0099).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We resolve this problem by inserting an entry with probability set to an otherwise-invalid value (\u2212oc).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Finally, since non-projective constructions often involve long-distance dependencies, the problem is closely related to the recovery of empty categories and non-local dependencies in constituency-based parsing (Johnson, 2002; Dienes and Dubey, 2003; Jijkoun and de Rijke, 2004; Cahill et al., 2004; Levy and Manning, 2004; Campbell, 2004).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"7).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Interpolation search formalizes the notion that one opens a dictionary near the end to find the word \u201czebra.\u201d Initially, the algorithm knows the array begins at b +\u2014 0 and ends at e +\u2014 |A|\u22121.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"During coreference resolution, the caseframe network provides evidence that an anaphor and prior noun phrase might be coreferent.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Using the terminology of Kahane et al. (1998), we say that jedna is the syntactic head of Z, while je is its linear head in the projectivized representation.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We currently simulate this by crafting a WCFG and feeding it to BitPar.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"To be short, we omit the target words e; e0 in the formulation of the search hypotheses.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"On one end of the spectrum are clustering approaches that assign a single POS tag to each word type (Schutze, 1995; Lamar et al., 2010).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"segmentation (Table 2).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"This &quot;default&quot; feature type has 100% coverage (it is seen on every example) but a low, baseline precision.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In this paper, we will propose an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In order to train this error-correction model, he divided his training corpus into 5 portions of 20% each.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The complexity of the algorithm is O(E3 J2 2J), where E is the size of the target language vocabulary.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We developed a first version of annotation guidelines for co-reference in PCC (Gross 2003), which served as basis for annotating the core corpus but have not been empirically evaluated for inter-annotator agreement yet.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"When two partial hypotheses have equal state (including that of other features), they can be recombined and thereafter efficiently handled as a single packed hypothesis.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Recently, statistical NERs have achieved results that are comparable to hand-coded systems.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Of course, since the number of attested (phonemic) Mandarin syllables (roughly 1400, including tonal distinctions) is far smaller than the number of morphemes, it follows that a given syllable could in principle be written with any of several different hanzi, depending upon which morpheme is intended: the syllable zhongl could be lfl 'middle,''clock,''end,' or ,'loyal.'","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Realizing gains in practice can be challenging, however, particularly when the target domain is distant from the background data.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"to explore how well we can induce POS tags using only the one-tag-per-word constraint.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Our work is motivated by the observation that contextual roles can be critically important in determining the referent of a noun phrase.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"It reads a file with a list of German connectives, and when a text is opened for annotation, it highlights all the words that show up in this list; these will be all the potential connectives.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"A defining characteristic of MSA is the prevalence of discourse markers to connect and subordinate words and phrases (Ryding, 2005).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The Levenshtein distance between the automatic translation and each of the reference translations is computed, and the minimum Levenshtein distance is taken.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"As is standard, we report the greedy one-to-one (Haghighi and Klein, 2006) and the many-to-one token-level accuracy obtained from mapping model states to gold POS tags.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Parameter Component As in the standard Bayesian HMM (Goldwater and Griffiths, 2007), all distributions are independently drawn from symmetric Dirichlet distributions: 2 Note that t and w denote tag and word sequences respectively, rather than individual tokens or tags.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"was done by the participants.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Thus in an English sentence such as I'm going to show up at the ACL one would reasonably conjecture that there are eight words separated by seven spaces.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The AdaBoost algorithm was developed for supervised learning.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We present a coreference resolver called BABAR that uses contextual role knowledge to evaluate possible antecedents for an anaphor.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The phrases have to be the expressions of length less than 5 chunks, appear between two NEs.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"A total of 13,976 phrases were grouped.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Note that while the standard HMM, has O(K n) emission parameters, our model has O(n) effective parameters.3 Token Component Once HMM parameters (\u00cf\u0086, \u00ce\u00b8) have been drawn, the HMM generates a token-level corpus w in the standard way: P (w, t|\u00cf\u0086, \u00ce\u00b8) = P (T , W , \u00ce\u00b8, \u00cf\u0088, \u00cf\u0086, t, w|\u00ce\u00b1, \u00ce\u00b2) = P (T , W , \u00cf\u0088|\u00ce\u00b2) [Lexicon] \u00ef\u00a3\u00ab n n \u00ef\u00a3\u00ad (w,t)\u00e2\u0088\u0088(w,t) j \u00ef\u00a3\u00b6 P (tj |\u00cf\u0086tj\u00e2\u0088\u00921 )P (wj |tj , \u00ce\u00b8tj )\u00ef\u00a3\u00b8 P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) [Parameter] P (w, t|\u00cf\u0086, \u00ce\u00b8) [Token] We refer to the components on the right hand side as the lexicon, parameter, and token component respectively.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"These 140 NE categories are designed by extending MUC\u00e2\u0080\u0099s 7 NE categories with finer sub-categories (such as Company, Institute, and Political Party for Organization; and Country, Province, and City for Location) and adding some new types of NE categories (Position Title, Product, Event, and Natural Object).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"5 64.7 42.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"(2009) study related but different multilingual grammar and tagger induction tasks, where it is assumed that no labeled data at all is available.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Then, moving from connective to connective, ConAno sometimes offers suggestions for its scope (using heuristics like \u00e2\u0080\u0098for sub- junctor, mark all words up to the next comma as the first segment\u00e2\u0080\u0099), which the annotator can accept with a mouseclick or overwrite, marking instead the correct scope with the mouse.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The price shocked her.), or same-kind (e.g., Her health insurance paid for the hospital fees, but the automobile insurance did not cover the repair.).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"4 69.0 51.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Taking \/V, to be the number of examples an algorithm classified correctly (where all gold standard items labeled noise were counted as being incorrect), we calculated two measures of accuracy: See Tab.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The method just described segments dictionary words, but as noted in Section 1, there are several classes of words that should be handled that are not found in a standard dictionary.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Having found (spelling, context) pairs in the parsed data, a number of features are extracted.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"of Tokens No.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We hope that this will allow practitioners to apply our approach directly to languages for which no resources are available.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In general, m; l; l0 6= fl1; l2; l3g and in line umber 3 and 4, l0 must be chosen not to violate the above reordering restriction.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Formally, for a lexicon L and segments I \u00e2\u0088\u0088 L, O \u00e2\u0088\u0088\/ L, each word automaton accepts the language I\u00e2\u0088\u0097(O + I)I\u00e2\u0088\u0097.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"GL is then used to parse the string tn1 ... tnk_1, where tni is a terminal corresponding to the lattice span between node ni and ni+1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This WFST represents the segmentation of the text into the words AB and CD, word boundaries being marked by arcs mapping between f and part-of-speech labels.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We have modified Moses (Koehn et al., 2007) to keep our state with hypotheses; to conserve memory, phrases do not keep state.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Further, the special hash 0 suffices to flag empty buckets.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"2.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"was done by the participants.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Text generation, or at least the two phases of text planning and sentence planning, is a process driven partly by well-motivated choices (e.g., use this lexeme X rather than that more colloquial near-synonym Y ) and partly by con tation like that of PCC can be exploited to look for correlations in particular between syntactic structure, choice of referring expressions, and sentence-internal information structure.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This has the potential drawback of increasing the number of features, which can make MERT less stable (Foster and Kuhn, 2009).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We checked whether the discovered links are listed in WordNet.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Pr(eI 1) is the language model of the target language, whereas Pr(fJ 1 jeI1) is the transla tion model.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"This is the form of recursive levels in iDafa constructs.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"So, 1: f, xue2shengl+men0 (student+PL) 'students' occurs and we estimate its cost at 11.43; similarly we estimate the cost of f, jiang4+men0 (general+PL) 'generals' (as in 'J' f, xiao3jiang4+men0 'little generals'), at 15.02.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The key point is that the second constraint can be remarkably powerful in reducing the complexity of the learning problem.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Feature weights were set using Och\u2019s MERT algorithm (Och, 2003).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"It also incorporates the Good-Turing method (Baayen 1989; Church and Gale 1991) in estimating the likelihoods of previously unseen con\u00c2\u00ad structions, including morphological derivatives and personal names.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Here NO counts the number of hypothesized constituents in the development set that match the binary predicate specified as an argument.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"shortest match at each point.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Roughly speaking, previous work can be divided into three categories, namely purely statistical approaches, purely lexi\u00c2\u00ad cal rule-based approaches, and approaches that combine lexical information with sta\u00c2\u00ad tistical information.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"was done by the participants.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The weak hypothesis can abstain from predicting the label of an instance x by setting h(x) = 0.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"At present, the \u00e2\u0080\u0098Potsdam Commentary Corpus\u00e2\u0080\u0099 (henceforth \u00e2\u0080\u0098PCC\u00e2\u0080\u0099 for short) consists of 170 commentaries from Ma\u00c2\u00a8rkische Allgemeine Zeitung, a German regional daily.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Across eight European languages, our approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"All commentaries have been annotated with rhetorical structure, using RSTTool4 and the definitions of discourse relations provided by Rhetorical Structure Theory (Mann, Thompson 1988).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We often encounter sentences that are highly ambiguous in themselves, without some prior knowledge of the entities concerned.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The confidence level is then used as the belief value for the knowledge source.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"In all cases, the key is collapsed to its 64-bit hash.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Later, BerkeleyLM (Pauls and Klein, 2011) described ideas similar to ours.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Our final average POS tagging accuracy of 83.4% compares very favorably to the average accuracy of Berg-Kirkpatrick et al.\u2019s monolingual unsupervised state-of-the-art model (73.0%), and considerably bridges the gap to fully supervised POS tagging performance (96.6%).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The secondary classifier in (Borthwick, 1999) uses information not just from the current article, but also from the whole test corpus, with an additional feature that indicates if the information comes from the same document or from another document.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This is orthographically represented as 7C.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The constituent voting and na\u00efve Bayes techniques are equivalent because the parameters learned in the training set did not sufficiently discriminate between the three parsers.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"An input ABCD can be represented as an FSA as shown in Figure 2(b).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"This paper does not necessarily reflect the position of the U.S. Government.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"64 94.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Less frequently studied is the interplay among language, annotation choices, and parsing model design (Levy and Manning, 2003; Ku\u00c2\u00a8 bler, 2005).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"More examples are shown in Figure 5.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"A direct-mapped cache makes BerkeleyLM faster on repeated queries, but their fastest (scrolling) cached version is still slower than uncached PROBING, even on cache-friendly queries.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"4 7 . 3 8 . 9 2 8 . 8 2 0 . 7 3 2 . 3 3 5 . 2 2 9 . 6 2 7 . 6 1 4 . 2 4 2 . 8 4 5 . 9 4 4 . 3 6 0 . 6 6 1 . 5 4 9 . 9 3 3 . 9 Table 6: Type-level Results: Each cell report the type- level accuracy computed against the most frequent tag of each word type.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"However, supervised methods rely on labeled training data, which is time-consuming and expensive to generate.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"In IdentiFinder, there is a priority in the feature assignment, such that if one feature is used for a token, another feature lower in priority will not be used.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We can define derivation trees inductively on the length of the derivation of a tree 1.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The Levenshtein distance between the automatic translation and each of the reference translations is computed, and the minimum Levenshtein distance is taken.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Finally, we intend to explore more sophisticated instanceweighting features for capturing the degree of generality of phrase pairs.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Nu mb er filters candidate if number doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"It is worth noting that, although nonprojective constructions are less frequent in DDT than in PDT, they seem to be more deeply nested, since only about 80% can be projectivized with a single lift, while almost 95% of the non-projective arcs in PDT only require a single lift.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"And time is short.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The first shows how constituent features and context do not help in deciding which parser to trust.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"collected too.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The matching sentence pairs are then added to the IN corpus, and the system is re-trained.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The learning task is to find two classifiers : 2x1 { \u20141, +1} 12 : 2x2 { \u20141, +1} such that (x1,) = f2(x2,t) = yt for examples i = 1, , m, and f1 (x1,) = f2 (x2,t) as often as possible on examples i = m + 1, ,n. To achieve this goal we extend the auxiliary function that bounds the training error (see Equ.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"5 Related Work.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Roughly speaking, previous work can be divided into three categories, namely purely statistical approaches, purely lexi\u00c2\u00ad cal rule-based approaches, and approaches that combine lexical information with sta\u00c2\u00ad tistical information.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Mutual information was shown to be useful in the segmentation task given that one does not have a dictionary.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"We apply a beam search concept as in speech recognition.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Link phrases based on instance pairs Using NE instance pairs as a clue, we find links between sets of phrases.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"However, the learning curves in Figure 3 show that the Berkeley parser does not exceed our manual grammar by as wide a margin as has been shown for other languages (Petrov, 2009).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The type-level tag assignments T generate features associated with word types W . The tag assignments constrain the HMM emission parameters \u00ce\u00b8.","label":0},{"source_text":"There is no global pruning.","target_text":"(f1; ;mg n fl1g ; l) 3 (f1; ;mg n fl; l1; l2g ; l0) !","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"All four of the techniques studied result in parsing systems that perform better than any previously reported.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"2 Chinese ?l* han4zi4 'Chinese character'; this is the same word as Japanese kanji..","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"To optimize left-to-right queries, we extend state to store backoff information: where m is the minimal context from Section 4.1 and b is the backoff penalty.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In our third model GTppp we also add the distinction between general PPs and possessive PPs following Goldberg and Elhadad (2007).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We can model this probability straightforwardly enough with a probabilistic version of the grammar just given, which would assign probabilities to the individual rules.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"There exist a few robust broad-coverage parsers that produce non-projective dependency structures, notably Tapanainen and J\u00a8arvinen (1997) and Wang and Harper (2004) for English, Foth et al. (2004) for German, and Holan (2004) for Czech.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"It was developed in response to the non-terminal\/terminal bias of Evalb, but Clegg and Shepherd (2005) showed that it is also a valuable diagnostic tool for trees with complex deep structures such as those found in the ATB.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(See Sproat and Shih 1995.)","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This is not to say that a set of standards by which a particular segmentation would count as correct and another incorrect could not be devised; indeed, such standards have been proposed and include the published PRCNSC (1994) and ROCLING (1993), as well as the unpublished Linguistic Data Consortium standards (ca.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We evaluated the results based on two metrics.","label":0},{"source_text":"There is no global pruning.","target_text":"The translation search is carried out with the category markers and the city names are resubstituted into the target sentence as a postprocessing step.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"It is unique, agrees with the maximum-likelihood distribution, and has the exponential form (Della Pietra et al., 1997): where refers to the outcome, the history (or context), and is a normalization function.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The role that each noun phrase plays in the kidnapping event is key to distinguishing these cases.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Maamouri et al.","label":0},{"source_text":"There is no global pruning.","target_text":"A summary of the corpus used in the experiments is given in Table 3.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In this paper k = 3 (the three labels are person, organization, location), and we set a = 0.1.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Compared to the EMEA\/EP setting, the two domains in the NIST setting are less homogeneous and more similar to each other; there is also considerably more IN text available.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"This research was supported in part by the Defense Advanced Research Projects Agency as part of the Translingual Information Detection, Extraction and Summarization (TIDES) program, under Grant N66001001-18917 from the Space and Naval Warfare Systems Center, San Diego, and by the National Science Foundation under Grant IIS00325657.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Word type N % Dic tion ary entr ies 2 , 5 4 3 9 7 . 4 7 Mor pho logi call y deri ved wor ds 3 0 . 1 1 Fore ign tran slite rati ons 9 0 . 3 4 Per son al na mes 5 4 2 . 0 7 cases.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"For each experiment we gave an nonparametric and a parametric technique for combining parsers.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Machine Translation In this case my colleague can not visit on I n d i e s e m F a l l ka nn m e i n K o l l e g e a m the v i e r t e n M a i n i c h t b e s u c h e n S i e you fourth of May Figure 1: Reordering for the German verbgroup.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Historically, Arabic grammar has identified two sentences types: those that begin with a nominal (\u00ef\u00bf\u00bd '.i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We would like to thank Prof. Ralph Grish- man, Mr. Takaaki Hasegawa and Mr. Yusuke Shinyama for useful comments, discussion and evaluation.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"basically complete, yet some improvements and extensions are still under way.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"G1 and G2 are hanzi, we can estimate the probability of the sequence being a name as the product of: \u00e2\u0080\u00a2 the probability that a word chosen randomly from a text will be a name-p(rule 1), and \u00e2\u0080\u00a2 the probability that the name is of the form 1hanzi-family 2hanzi-given-p(rule 2), and \u00e2\u0080\u00a2 the probability that the family name is the particular hanzi F1-p(rule 6), and \u00e2\u0080\u00a2 the probability that the given name consists of the particular hanzi G1 and G2-p(rule 9) This model is essentially the one proposed in Chang et al.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Prague Dependency Treebank (Haji\u02c7c et al., 2001b), Danish Dependency Treebank (Kromann, 2003), and the METU Treebank of Turkish (Oflazer et al., 2003), which generally allow annotations with nonprojective dependency structures.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"97 78.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"First, it directly encodes linguistic intuitions about POS tag assignments: the model structure reflects the one-tag-per-word property, and a type- level tag prior captures the skew on tag assignments (e.g., there are fewer unique determiners than unique nouns).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"MENE is then trained on 80% of the training corpus, and tested on the remaining 20%.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"MADA uses an ensemble of SVMs to first re-rank the output of a deterministic morphological analyzer.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The foreign language vertices (denoted by Vf) correspond to foreign trigram types, exactly as in Subramanya et al. (2010).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"For example, we might have VP \u00e2\u0086\u0092 VB NP PP, where the NP is the subject.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For a language like English, this problem is generally regarded as trivial since words are delimited in English text by whitespace or marks of punctuation.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"One of the strengths of the DempsterShafer model is its natural ability to recognize when several credible hypotheses are still in play.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"One way to approach this discrepancy is to assume a preceding phase of morphological segmentation for extracting the different lexical items that exist at the token level (as is done, to the best of our knowledge, in all parsing related work on Arabic and its dialects (Chiang et al., 2006)).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For example, if {N P1, N P2, N P3} are all coreferent, then each NP must be linked to one of the other two NPs.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Once the lexicon has been drawn, the model proceeds similarly to the standard token-level HMM: Emission parameters \u00ce\u00b8 are generated conditioned on tag assignments T . We also draw transition parameters \u00cf\u0086.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Maximizing (7) is thus much faster than a typical MERT run. where co(s, t) are the counts from OUT, as in (6).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Finally, we wish to reiterate an important point.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(b) 89 :1 t& tal de cai2neng2 hen3 he DE talent very 'He has great talent' f.b ga ol hig h While the current algorithm correctly handles the (b) sentences, it fails to handle the (a) sentences, since it does not have enough information to know not to group the sequences.ma3lu4 and?]cai2neng2 respectively.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"On several languages, we report performance exceeding that of more complex state-of-the art systems.1","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We have presented two general approaches to studying parser combination: parser switching and parse hybridization.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For example, syntactic decoders (Koehn et al., 2007; Dyer et al., 2010; Li et al., 2009) perform dynamic programming parametrized by both backward- and forward-looking state.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Table 2 shows BABAR\u00e2\u0080\u0099s performance.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"3 68.4 49.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The 1st block contains the simple baselines from section 2.1.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"However, it is unlikely that other occurrences of News Broadcasting Corp. in the same document also co-occur with Even.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Timing is based on plentiful memory.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"F-measure is the harmonic mean of precision and recall, 2PR\/(P + R).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Each spawned process must check if xi , , xn, and , yn, can be derived from B and C, respectively.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In our third model GTppp we also add the distinction between general PPs and possessive PPs following Goldberg and Elhadad (2007).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"A graphical depiction of our model as well as a summary of random variables and parameters can be found in Figure 1.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"It is interesting to note, however, that the ability to produce a bounded number of dependent paths (where two dependent paths can share an unbounded amount of information) does not require machinery as powerful as that used in LFG, FUG and IG's.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The paper is organized as follows: Section 2 explains the different layers of annotation that have been produced or are being produced.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Our empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","label":1},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Thus an explicit assumption about the redundancy of the features \u2014 that either the spelling or context alone should be sufficient to build a classifier \u2014 has been built into the algorithm.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"2 for the accuracy of the different methods.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"As in the case of the derivation trees of CFG's, nodes are labeled by a member of some finite set of symbols (perhaps only implicit in the grammar as in TAG's) used to denote derived structures.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"This algorithm can be applied to statistical machine translation.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"To set \u03b2, we used the same criterion as for \u03b1, over a dev corpus: The MAP combination was used for TM probabilities only, in part due to a technical difficulty in formulating coherent counts when using standard LM smoothing techniques (Kneser and Ney, 1995).3 Motivated by information retrieval, a number of approaches choose \u201crelevant\u201d sentence pairs from OUT by matching individual source sentences from IN (Hildebrand et al., 2005; L\u00a8u et al., 2007), or individual target hypotheses (Zhao et al., 2004).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In this paper, we offer broad insight into the underperformance of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"2.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"An easy way to achieve this is to put the domain-specific LMs and TMs into the top-level log-linear model and learn optimal weights with MERT (Och, 2003).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This heuristics is used to prune all segmentation possibilities involving \u201clexically improper\u201d segments.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Local features are features that are based on neighboring tokens, as well as the token itself.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Denote the unthresholded classifiers after t \u2014 1 rounds by git\u20141 and assume that it is the turn for the first classifier to be updated while the second one is kept fixed.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"A contextual role represents the role that a noun phrase plays in an event or relationship.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"In our model, however, all lattice paths are taken to be a-priori equally likely.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"\u00e2\u0080\u0094ested in correlations between prosody and dis course structure.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"However, the learning curves in Figure 3 show that the Berkeley parser does not exceed our manual grammar by as wide a margin as has been shown for other languages (Petrov, 2009).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"3 68.4 49.","label":0},{"source_text":"There is no global pruning.","target_text":"The search starts in hypothesis (f;g; 0) and ends in the hypotheses (f1; ; Jg; j), with j 2 f1; ; Jg.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If \u00e2\u0080\u009cgun\u00e2\u0080\u009d and \u00e2\u0080\u009crevolver\u00e2\u0080\u009d refer to the same object, then it should also be acceptable to say that Fred was \u00e2\u0080\u009ckilled with a gun\u00e2\u0080\u009d and that the burglar \u00e2\u0080\u009cfireda revolver\u00e2\u0080\u009d.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This essentially means that a better grammar tunes the joint model for optimized syntactic disambiguation at least in as much as their hyper parameters do.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For all lists except locations, the lists are processed into a list of tokens (unigrams).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Its success depends on the two domains being relatively close, and on the OUT corpus not being so large as to overwhelm the contribution of IN.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The aforementioned surface form bcl, for example, may also stand for the lexical item \u201conion\u201d, a Noun.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We hypothesize that modeling morphological information will greatly constrain the set of possible tags, thereby further refining the representation of the tag lexicon.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The combining technique must act as a multi-position switch indicating which parser should be trusted for the particular sentence.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The results, along with the total number of phrases, are shown in Table 1.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(We would like to note though that unlike previous boosting algorithms, the CoBoost algorithm presented here is not a boosting algorithm under Valiant's (Valiant 84) Probably Approximately Correct (PAC) model.)","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"To optimize this function, we used L-BFGS, a quasi-Newton method (Liu and Nocedal, 1989).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"As mentioned above, it is not obvious how to apply Daum\u00b4e\u2019s approach to multinomials, which do not have a mechanism for combining split features.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Hence we decided to select ten commentaries to form a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099, for which the entire range of annotation levels was realized, so that experiments with multi-level querying could commence.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In the named entity task, X1 might be the instance space for the spelling features, X2 might be the instance space for the contextual features.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In the following, we assume that this word joining has been carried out.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"1.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The belief value that would have been assigned to the intersection of these sets is .60*.70=.42, but this belief has nowhere to go because the null set is not permissible in the model.7 So this probability mass (.42) has to be redistributed.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We then discuss how we adapt and generalize a boosting algorithm, AdaBoost, to the problem of named entity classification.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(Brin 98) ,describes a system for extracting (author, book-title) pairs from the World Wide Web using an approach that bootstraps from an initial seed set of examples.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Ltd., then organization will be more probable.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"To conserve memory at the expense of accuracy, values may be quantized using q bits per probability and r bits per backoff6.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In our coreference resolver, we define \u00ce\u00b8 to be the set of all candidate antecedents for an anaphor.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The computation of Pfr1(c)1Mi M k (C)) has been sketched before in Equations 1 through 4.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Given a key k, it estimates the position If the estimate is exact (A[pivot] = k), then the algorithm terminates succesfully.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The only way to handle such phenomena within the framework described here is simply to expand out the reduplicated forms beforehand, and incorporate the expanded forms into the lexical transducer.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"3 58.3 40.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"73 81.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"To resolve the anaphor, we survey the final belief values assigned to each candidate\u00e2\u0080\u0099s singleton set.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We again adopt an approach where we alternate between two classifiers: one classifier is modified while the other remains fixed.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In this paper, our goal has been to use the notion of LCFRS's to classify grammatical systems on the basis of their strong generative capacity.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Tables 4 and 5 also show that putting all of the contextual role KSs in play at the same time produces the greatest performance gain.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In approaching this problem, a variety of different methods are conceivable, including a more or less sophisticated use of machine learning.","label":0},{"source_text":"There is no global pruning.","target_text":"The advantage is that we can recombine search hypotheses by dynamic programming.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For the seen word ir, 'gen\u00c2\u00ad erals,' there is an c:NC transduction from to the node preceding ir,; this arc has cost cost( f,) - cost(unseen(f,)), so that the cost of the whole path is the desired cost( f,).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"First, we aim to explicitly characterize examples from OUT as belonging to general language or not.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Church and Hanks [1989]), and we have used lists of character pairs ranked by mutual information to expand our own dictionary.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"We have shown that the maximum entropy framework is able to use global information directly.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"3.4 Salience-based text generation.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Based on these high-confidence alignments we can extract tuples of the form [u H v], where u is a foreign trigram type, whose middle word aligns to an English word type v. Our bilingual similarity function then sets the edge weights in proportion to these tuple counts.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The complexity of the quasimonotone search is O(E3 J (R2+LR)).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"A related point is that mutual information is helpful in augmenting existing electronic dictionaries, (cf.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"\"c' 0 + 0 \"0 ' \u00e2\u0080\u00a2 + a n t i g r e e d y x g r e e d y < > c u r r e n t m e t h o d o d i e t . o n l y \u00e2\u0080\u00a2 Taiwan 0 \u00c2\u00b7;; 0 c CD E i5 0\"' 9 9 \u00e2\u0080\u00a2 Mainland \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 -0.30.20.1 0.0 0.1 0.2 Dimension 1 (62%) Figure 7 Classical metric multidimensional scaling of distance matrix, showing the two most significant dimensions.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Test set OOV rate is computed using the following splits: ATB (Chiang et al., 2006); CTB6 (Huang and Harper, 2009); Negra (Dubey and Keller, 2003); English, sections 221 (train) and section 23 (test).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This departure from the traditional token-based tagging approach allows us to explicitly capture type- level distributional properties of valid POS tag as signments as part of the model.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"See Table 2 for the tag set size of other languages.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The terms on the right-hand-side denote the type-level and token-level probability terms respectively.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"A cell in the bottom row of the parse chart is required for each potential whitespace boundary.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In that work, mutual information was used to decide whether to group adjacent hanzi into two-hanzi words.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The performance of our system on those sentences ap\u00c2\u00ad peared rather better than theirs.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The alignment model uses two kinds of parameters: alignment probabilities p(aj jaj\u00f4\u0080\u0080\u00801; I; J), where the probability of alignment aj for position j depends on the previous alignment position aj\u00f4\u0080\u0080\u00801 (Ney et al., 2000) and lexicon probabilities p(fj jeaj ).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Two issues distinguish the various proposals.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"2 62.6 45.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"8 1 8.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The cost estimate, cost(i\u00c2\u00a5JJ1l.fn is computed in the obvious way by summing the negative log probabilities of i\u00c2\u00a5JJ1l.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Therefore, for n-gram wn1 , all leftward extensions wn0 are an adjacent block in the n + 1-gram array.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"For all languages, the vocabulary sizes increase by several thousand words.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"It also incorporates the Good-Turing method (Baayen 1989; Church and Gale 1991) in estimating the likelihoods of previously unseen con\u00c2\u00ad structions, including morphological derivatives and personal names.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For all other recursive NPs, we add a common annotation to the POS tag of the head (recursiveNPHead).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Besides information structure, the second main goal is to enhance current models of rhetorical structure.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Arabic sentences of up to length 63 would need to be.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"While it is possible to derive a closed form solution for this convex objective function, it would require the inversion of a matrix of order |Vf|.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Such sequences are given additional features of A begin, A continue, or A end, and the acronym is given a feature A unique.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, a story can mention \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d, \u00e2\u0080\u009cthe White House\u00e2\u0080\u009d, or \u00e2\u0080\u009cthe weather\u00e2\u0080\u009d without any prior referent in the story.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Since our goal is to perform well under these measures we will similarly treat constituents as the minimal substructures for combination.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"1 1 0.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We were intentionally lenient with our baselines: bilingual information by projecting POS tags directly across alignments in the parallel data.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For that application, at a minimum, one would want to know the phonological word boundaries.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"(f1; ;mg ; l) 2 (f1; ;mg n fl; l1g ; l0) !","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"0 70.9 42.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"9 65.5 46.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"As already pointed out in Section 2.4, current theories diverge not only on the number and definition of relations but also on apects of structure, i.e., whether a tree is sufficient as a representational device or general graphs are required (and if so, whether any restrictions can be placed on these graph\u00e2\u0080\u0099s structures \u00e2\u0080\u0094 cf.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We suspect that the higher precision in the disasters domain may be due to its substantially larger training corpus.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The system described in this paper is similar to the MENE system of (Borthwick, 1999).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"On the other hand, only very restricted reorderings are necessary, e.g. for the translation direction from Table 2: Coverage set hypothesis extensions for the IBM reordering.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Many statistical or machine-learning approaches for natural language problems require a relatively large amount of supervision, in the form of labeled training examples.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For example: McCann initiated a new global system.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The general idea for the knowledge- based part is to have the system use as much information as it can find at its disposal to produce a target representation as specific as possible and as underspecified as necessary.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"As far as we know, no other NERs have used information from the whole document (global) as well as information within the same sentence (local) in one framework.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"It may be more realistic to replace the second criteria with a softer one, for example (Blum and Mitchell 98) suggest the alternative Alternatively, if Ii and 12 are probabilistic learners, it might make sense to encode the second constraint as one of minimizing some measure of the distance between the distributions given by the two learners.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The use of global features has improved the performance on MUC6 test data from 90.75% to 93.27% (27% reduction in errors), and the performance on MUC7 test data from 85.22% to 87.24% (14% reduction in errors).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"In graph-based learning approaches one constructs a graph whose vertices are labeled and unlabeled examples, and whose weighted edges encode the degree to which the examples they link have the same label (Zhu et al., 2003).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Linear weights are difficult to incorporate into the standard MERT procedure because they are \u201chidden\u201d within a top-level probability that represents the linear combination.1 Following previous work (Foster and Kuhn, 2007), we circumvent this problem by choosing weights to optimize corpus loglikelihood, which is roughly speaking the training criterion used by the LM and TM themselves.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Finally, model GTv = 2 includes parent annotation on top of the various state-splits, as is done also in (Tsarfaty and Sima\u2019an, 2007; Cohen and Smith, 2007).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The first stage identifies a keyword in each phrase and joins phrases with the same keyword into sets.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Unique Occurrences and Zone (UNIQ): This group of features indicates whether the word is unique in the whole document.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"7 www.cis.upenn.edu\/\u00e2\u0088\u00bcpdtb\/ 8 www.eml-research.de\/english\/Research\/NLP\/ Downloads had to buy a new car.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"This drastic tree manipulation is not appropriate for situations in which we want to assign particular structures to sentences.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We call this technique constituent voting.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"MITLM 0.4 (Hsu and Glass, 2008) is mostly designed for accurate model estimation, but can also compute perplexity.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"F1 85 Berkeley 80 Stanford.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The alignment model uses two kinds of parameters: alignment probabilities p(aj jaj\u00f4\u0080\u0080\u00801; I; J), where the probability of alignment aj for position j depends on the previous alignment position aj\u00f4\u0080\u0080\u00801 (Ney et al., 2000) and lexicon probabilities p(fj jeaj ).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Ends with the feminine affix :: p. 4.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Any NLP application that presumes as input unrestricted text requires an initial phase of text analysis; such applications involve problems as diverse as machine translation, information retrieval, and text-to-speech synthesis (TIS).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The 13 positions of the source sentence are processed in the order shown.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The taggers were trained on datasets labeled with the universal tags.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Also, in Information Extraction (IE), in which the system tries to extract elements of some events (e.g. date and company names of a corporate merger event), several event instances from different news articles have to be aligned even if these are expressed differently.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Each visited entry wni stores backoff b(wni ).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"However, our full model takes advantage of word features not present in Grac\u00c2\u00b8a et al.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"This approach leads to a search procedure with complexity O(E3 J4).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In this work, we take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The authors acknowledge the support of the NSF (CAREER grant IIS0448168, and grant IIS 0904684).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Next, for each pair of NE categories, we collect all the contexts and find the keywords which are topical for that NE category pair.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Instead, the designs of the various annotation layers and the actual annotation work are results of a series of diploma theses, of students\u00e2\u0080\u0099 work in course projects, and to some extent of paid assistentships.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In considering recognition of LCFRS's, we make further assumption concerning the contribution of each structure to the input string, and how the composition operations combine structures in this respect.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Acronyms (ACRO): Words made up of all capitalized letters in the text zone will be stored as acronyms (e.g., IBM).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Methods that allow multiple segmentations must provide criteria for choosing the best segmentation.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"If the context wnf will never extend to the right (i.e. wnf v is not present in the model for all words v) then no subsequent query will match the full context.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"40 75.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Following the setup of Johnson (2007), we use the whole of the Penn Treebank corpus for training and evaluation on English.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Vocabulary lookup is a sorted array of 64-bit word hashes.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"3.2 Stochastic rhetorical analysis.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"can expect famous names like Zhou Enlai's to be in many dictionaries, but names such as :fi lf;f; shi2jil-lin2, the name of the second author of this paper, will not be found in any dictionary.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Phrase-level granularity distinguishes our work from previous work by Matsoukas et al (2009), who weight sentences according to sub-corpus and genre membership.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The first method uses a similar algorithm to that of (Yarowsky 95), with modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Somewhat surprisingly, there do not appear to be large systematic differences between linear and MAP combinations.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"From this we see that a finer-grained model for parser combination, at least for the features we have examined, will not give us any additional power.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We call this technique constituent voting.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"For this reason, na\u00efve Bayes classifiers are well-matched to this problem.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Words and punctuation that appear in brackets are considered optional.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In (1) the sequencema3lu4 cannot be resolved locally, but depends instead upon broader context; similarly in (2), the sequence :::tcai2neng2 cannot be resolved locally: 1.","label":0},{"source_text":"There is no global pruning.","target_text":"4.1 The Task and the Corpus.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each domain, we created a blind test set by manually annotating 40 doc uments with anaphoric chains, which represent sets of m3 (S) = ) X \u00e2\u0088\u00a9Y =S 1 \u00e2\u0088\u0092 ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) (1) noun phrases that are coreferent (as done for MUC6 (MUC6 Proceedings, 1995)).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"However, it is desirable if we can separate them.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"While Berg-Kirkpatrick et al.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold, 30.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Moreover, the Stanford parser achieves the most exact Leaf Ancestor matches and tagging accuracy that is only 0.1% below the Bikel model, which uses pre-tagged input.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The extent to which this constraint is enforced varies greatly across existing methods.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Gazdar (1985) argues that sharing of stacks can be used to give analyses for coordination.","label":0},{"source_text":"There is no global pruning.","target_text":"In Section 2, we brie y review our approach to statistical machine translation.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Hence, s < k. But by addition of the votes on the two parses, s > 2N-11> k, a contradiction.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"It is the performance we could achieve if an omniscient observer told us which parser to pick for each of the sentences.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We refer to this process as Reliable Case Resolution because it involves finding cases of anaphora that can be easily resolved with their antecedents.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"First we mark any node that dominates (at any level) a verb sider POS tags when pre-terminals are the only intervening nodes between the nucleus and its bracketing (e.g., unaries, base NPs).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Hence we decided to restrict ourselves to only information from the same document.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"f, nan2gual+men0 'pumpkins' is by no means impossible.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"However, their system is a hybrid of hand-coded rules and machine learning methods.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Our TRIE implementation is designed to improve upon IRSTLM using a reverse trie with improved search, bit level packing, and stateful queries.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In MSA, SVO usually appears in non-matrix clauses.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The data is sorted based on the frequency of the context (\u00e2\u0080\u009ca unit of\u00e2\u0080\u009d appeared 314 times in the corpus) and the NE pair instances appearing with that context are shown with their frequency (e.g. \u00e2\u0080\u009cNBC\u00e2\u0080\u009d and \u00e2\u0080\u009cGeneral Electric Co.\u00e2\u0080\u009d appeared 10 times with the context \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Indeed, as we shall show in Section 5, even human judges differ when presented with the task of segmenting a text into words, so a definition of the criteria used to determine that a given segmentation is correct is crucial before one can interpret such measures.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In our model, we associate these features at the type-level in the lexicon.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For derived words that occur in our corpus we can estimate these costs as we would the costs for an underived dictionary entry.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Sentence (2) and (3) help to disambiguate one way or the other.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"36.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"First, we identify sources of syntactic ambiguity understudied in the existing parsing literature.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The first m pairs have labels yi, whereas for i = m + 1, , n the pairs are unlabeled.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Morphological processes in Semitic languages deliver space-delimited words which introduce multiple, distinct, syntactic units into the structure of the input sentence.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The 1-bit sign is almost always negative and the 8-bit exponent is not fully used on the range of values, so in practice this corresponds to quantization ranging from 17 to 20 total bits.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Table 9: Dev set results for sentences of length \u00e2\u0089\u00a4 70.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For example, as Gan (1994) has noted, one can construct examples where the segmen\u00c2\u00ad tation is locally ambiguous but can be determined on the basis of sentential or even discourse context.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"A morpheme, on the other hand, usually corresponds to a unique hanzi, though there are a few cases where variant forms are found.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Linguistic intuitions like those in the previous section inform language-specific annotation choices.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The range of the score is between 0 and 1 (higher is better).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Besides the lack of a clear definition of what constitutes a correct segmentation for a given Chinese sentence, there is the more general issue that the test corpora used in these evaluations differ from system to system, so meaningful comparison between systems is rendered even more difficult.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Linear probing places at most one entry in each bucket.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For each co-occurrence relation (noun\/caseframe for CFLex, and caseframe\/caseframe for CFNet), BABAR computes its log-likelihood value and looks it up in the \u00cf\u00872 table to obtain a confidence level.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Examining the word fidanzato for the \u201cNo LP\u201d and \u201cWith LP\u201d models is particularly instructive.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The number of top-ranked pairs to retain is chosen to optimize dev-set BLEU score.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The restriction can be expressed in terms of the number of uncovered source sentence positions to the left of the rightmost position m in the coverage set.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The inventions disclosed herein are the subject of a patent application owned by the University of Utah and licensed on an exclusive basis to Attensity Corporation.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Our experiments consistently demonstrate that this model architecture yields substantial performance gains over more complex tagging counterparts.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"If two nouns have mutually exclusive semantic classes, then they cannot be coreferent.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"When the same token is to be interpreted as a single lexeme fmnh, it may function as a single adjective \u201cfat\u201d.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Either save money at any cost - or give priority to education.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The commentaries in PCC are all of roughly the same length, ranging from 8 to 10 sentences.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Figure 2: Order in which source positions are visited for the example given in Fig.1.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The overall performance of our joint framework demonstrates that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperforms upper bounds proposed by previous joint disambiguation systems and achieves segmentation and parsing results on a par with state-of-the-art standalone applications results.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Figure 2 shows examples of extracted NE pair instances and their contexts.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Using structural information As was explained in the results section, we extracted examples like \u00e2\u0080\u009cSmith estimates Lotus\u00e2\u0080\u009d, from a sentence like \u00e2\u0080\u009cMr.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The second case involves existential noun phrases (Allen, 1995), which are noun phrases that uniquely specify an object or concept and therefore do not need a prior referent in the discourse.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We report results for the best and median hyperparameter settings obtained in this way.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"8 We use head-finding rules specified by a native speaker.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"2.2.3 Lexical Caseframe Expectations The second type of contextual role knowledge learned by BABAR is Lexical Caseframe Expectations, which are used by the CFLex knowledge source.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In this paper we present a stochastic finite-state model for segmenting Chinese text into words, both words found in a (static) lexicon as well as words derived via the above-mentioned productive processes.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"To lower the barrier of entrance to the competition, we provided a complete baseline MT system, along with data resources.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Unlike the WSJ corpus which has a high frequency of rules like VP \u00e2\u0086\u0092VB PP, Arabic verb phrases usually have lexi calized intervening nodes (e.g., NP subjects and direct objects).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The monotone search performs worst in terms of both error rates mWER and SSER.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Morphological disambiguators that consider a token in context (an utterance) and propose the most likely morphological analysis of an utterance (including segmentation) were presented by Bar-Haim et al. (2005), Adler and Elhadad (2006), Shacham and Wintner (2007), and achieved good results (the best segmentation result so far is around 98%).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In addition, each feature function is a binary function.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The method being described-henceforth ST..","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The content does not necessarily reflect the views of the U.S. Government, and no official endorsement should be inferred.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The size of TRIE is particularly sensitive to F1092 c11, so vocabulary filtering is quite effective at reducing model size.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"(Thus the domain of the dev and test corpora matches IN.)","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For the seen word ir, 'gen\u00c2\u00ad erals,' there is an c:NC transduction from to the node preceding ir,; this arc has cost cost( f,) - cost(unseen(f,)), so that the cost of the whole path is the desired cost( f,).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"One side of the decision making process is when we choose to believe a constituent should be in the parse, even though only one parser suggests it.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"In the first stage, we run a single step of label propagation, which transfers the label distributions from the English vertices to the connected foreign language vertices (say, Vf\ufffd) at the periphery of the graph.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In this paper, Section 2 begins by explaining how contextual role knowledge is represented and learned.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Given this similarity function, we define a nearest neighbor graph, where the edge weight for the n most similar vertices is set to the value of the similarity function and to 0 for all other vertices.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"In (Reitter, Stede 2003) we went a different way and suggested URML5, an XML format for underspecifying rhetorical structure: a number of relations can be assigned instead of a single one, competing analyses can be represented with shared forests.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The second row represents the performance of the median hyperparameter setting.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(We would like to note though that unlike previous boosting algorithms, the CoBoost algorithm presented here is not a boosting algorithm under Valiant's (Valiant 84) Probably Approximately Correct (PAC) model.)","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The first modification \u2014 cautiousness \u2014 is a relatively minor change.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We compare three hash tables: our probing implementation, GCC\u2019s hash set, and Boost\u2019s8 unordered.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Maamouri et al.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This is an issue that we have not addressed at the current stage of our research.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Then each arc of D maps either from an element of H to an element of p, or from E-i.e., the empty string-to an element of P. More specifically, each word is represented in the dictionary as a sequence of arcs, starting from the initial state of D and labeled with an element 5 of Hxp, which is terminated with a weighted arc labeled with an element of Ex P. The weight represents the estimated cost (negative log probability) of the word.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"However, as we have noted, nothing inherent in the approach precludes incorporating higher-order constraints, provided they can be effectively modeled within a finite-state framework.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Also, the method of using keywords rules out phrases which don\u00e2\u0080\u0099t contain popular words in the domain.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Our original hope in combining these parsers is that their errors are independently distributed.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The same numbers were used for each data structure.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Our work is closest to that of Yarowsky and Ngai (2001), but differs in two important ways.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"In most cases, however, these expansions come with a steep increase in model complexity, with respect to training procedure and inference time.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"3 54.4 33.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"5.2 Discussion.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Modifying the Berkeley parser for Arabic is straightforward.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We can now add a new weak hypothesis 14 based on a feature in X1 with a confidence value al hl and atl are chosen to minimize the function We now define, for 1 <i <n, the following virtual distribution, As before, Ztl is a normalization constant.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For each token , zero, one, or more of the features in each feature group are set to 1.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(See also Wu and Fung [1994].)","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For a given \"word\" in the automatic segmentation, if at least k of the hu\u00c2\u00ad man judges agree that this is a word, then that word is considered to be correct.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Each visited entry wni stores backoff b(wni ).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"While possible to utilize the feature-based log-linear approach described in Berg-Kirkpatrick et al.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"A corpus of German newspaper commentaries has been assembled and annotated with different information (and currently, to different degrees): part-of-speech, syntax, rhetorical structure, connectives, co-reference, and information structure.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"This technique has the advantage of requiring no training, but it has the disadvantage of treating all parsers equally even though they may have differing accuracies or may specialize in modeling different phenomena.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For all variants, we found that BerkeleyLM always rounds the floating-point mantissa to 12 bits then stores indices to unique rounded floats.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"In these experiments, the input lacks segmentation markers, hence the slightly different dev set baseline than in Table 6.","label":0},{"source_text":"There is no global pruning.","target_text":"Can we do . QmS: Yes, wonderful.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"In the interest of testing the robustness of these combining techniques, we added a fourth, simple nonlexicalized PCFG parser.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Analysis of the data revealed that the contextual role knowledge is especially helpful for resolving pronouns because, in general, they are semantically weaker than definite NPs.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"During coreference resolution, BABAR checks (1) whether the anaphor is among the lexical expectations for the caseframe that extracts the candidate antecedent, and (2) whether the candidate is among the lexical expectations for the caseframe that extracts the anaphor.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The latter arcs correspond to OOV words in English.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"This departure from the traditional token-based tagging approach allows us to explicitly capture type- level distributional properties of valid POS tag as signments as part of the model.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Using this heuristic, BABAR identifies existential definite NPs in the training corpus using our previous learning algorithm (Bean and Riloff, 1999) and resolves all occurrences of the same existential NP with each another.1 2.1.2 Syntactic Seeding BABAR also uses syntactic heuristics to identify anaphors and antecedents that can be easily resolved.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"f, nan2gual+men0 'pumpkins' is by no means impossible.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"However, the point of RandLM is to scale to even larger data, compensating for this loss in quality.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"So the success of the algorithm may well be due to its success in maximizing the number of unlabeled examples on which the two decision lists agree.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Mutual information was shown to be useful in the segmentation task given that one does not have a dictionary.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Similar advances have been made in machine translation (Frederking and Nirenburg, 1994), speech recognition (Fiscus, 1997) and named entity recognition (Borthwick et al., 1998).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Particles are uninflected.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Our results show that MENERGI performs as well as IdentiFinder when trained on comparable amount of training data.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"1 1 0.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In this section, we try to compare our results with those obtained by IdentiFinder ' 97 (Bikel et al., 1997), IdentiFinder ' 99 (Bikel et al., 1999), and MENE (Borthwick, 1999).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As noted, this sentence consists of four words, namely B X ri4wen2 'Japanese,' :\u00c2\u00a5, zhanglyu2 'octopus\/ :&P:l zen3me0 'how,' and IDt shuol 'say.'","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Finally, this effort is part of a much larger program that we are undertaking to develop stochastic finite-state methods for text analysis with applications to TIS and other areas; in the final section of this paper we will briefly discuss this larger program so as to situate the work discussed here in a broader context.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"First, we describe how the caseframes are represented and learned.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We would like to thank Prof. Ralph Grish- man, Mr. Takaaki Hasegawa and Mr. Yusuke Shinyama for useful comments, discussion and evaluation.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"To optimize this function, we used L-BFGS, a quasi-Newton method (Liu and Nocedal, 1989).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Section 4 presents experimen tal results on two corpora: the MUC4 terrorism corpus, and Reuters texts about natural disasters.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"These systems rely on a training corpus that has been manually annotated with coreference links.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Feature-based HMM Model (Berg- Kirkpatrick et al., 2010): The KM model uses a variety of orthographic features and employs the EM or LBFGS optimization algorithm; Posterior regulariation model (Grac\u00c2\u00b8a et al., 2009): The G10 model uses the posterior regular- ization approach to ensure tag sparsity constraint.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The transition from f, to a final state transduces c to the grammatical tag PL with cost cost(unseen(f,)): cost(i\u00c2\u00a5JJ1l.ir,) == cost(i\u00c2\u00a5JJ1l.)","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Our oracles took advantage of the labeled treebanks: While we tried to minimize the number of free parameters in our model, there are a few hyperparameters that need to be set.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"A final alternate approach would be to combine weighted joint frequencies rather than conditional estimates, ie: cI(s, t) + w,\\(s, t)co(, s, t), suitably normalized.5 Such an approach could be simulated by a MAP-style combination in which separate 0(t) values were maintained for each t. This would make the model more powerful, but at the cost of having to learn to downweight OUT separately for each t, which we suspect would require more training data for reliable performance.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"TPT has theoretically better locality because it stores ngrams near their suffixes, thereby placing reads for a single query in the same or adjacent pages.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Riloff and Shepherd 97) describe a bootstrapping approach for acquiring nouns in particular categories (such as &quot;vehicle&quot; or &quot;weapon&quot; categories).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We received submissions from 14 groups from 11 institutions, as listed in Figure 2.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Since Dane\u00cb\u0087s\u00e2\u0080\u0099 proposals of \u00e2\u0080\u0098thematic development patterns\u00e2\u0080\u0099, a few suggestions have been made as to the existence of a level of discourse structure that would predict the information structure of sentences within texts.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"In Figure 4 we show an example of variation between the parsing models.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"They also describe an application of cotraining to classifying web pages (the to feature sets are the words on the page, and other pages pointing to the page).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Our second contribution is to apply instance weighting at the level of phrase pairs.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Since our graph is built from a parallel corpus, we can use standard word alignment techniques to align the English sentences De 5Note that many combinations are impossible giving a PMI value of 0; e.g., when the trigram type and the feature instantiation don\u2019t have words in common. and their foreign language translations Df.6 Label propagation in the graph will provide coverage and high recall, and we therefore extract only intersected high-confidence (> 0.9) alignments De\ufffdf.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In such cases we use the non-pruned lattice including all (possibly ungrammatical) segmentation, and let the statistics (including OOV) decide.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"i..f,..","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"To lower the barrier of entrance to the competition, we provided a complete baseline MT system, along with data resources.","label":0},{"source_text":"Here we present two algorithms.","target_text":".","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(Blum and Mitchell 98) go on to give PAC results for learning in the cotraining case.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This section describes an algorithm based on boosting algorithms, which were previously developed for supervised machine learning problems.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If evidence indicates that hypotheses C and D are less likely than hypotheses A and B, then probabilities are redistributed to reflect the fact that {A, B} is more likely to contain the answer than {C, D}.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, as we have noted, nothing inherent in the approach precludes incorporating higher-order constraints, provided they can be effectively modeled within a finite-state framework.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We call this approach parse hybridization.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"For Experiment 1 it is meaningless as a baseline, since it would result in 0% accuracy. mation on path labels but drop the information about the syntactic head of the lifted arc, using the label d\u2191 instead of d\u2191h (AuxP\u2191 instead of AuxP\u2191Sb).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"However, supervised methods rely on labeled training data, which is time-consuming and expensive to generate.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This suggests a strategy: run interpolation search until the range narrows to 4096 or fewer entries, then switch to binary search.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"96 75.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Since our graph is built from a parallel corpus, we can use standard word alignment techniques to align the English sentences De 5Note that many combinations are impossible giving a PMI value of 0; e.g., when the trigram type and the feature instantiation don\u2019t have words in common. and their foreign language translations Df.6 Label propagation in the graph will provide coverage and high recall, and we therefore extract only intersected high-confidence (> 0.9) alignments De\ufffdf.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The index in this array is the vocabulary identifier.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If any of the tokens from to is in Person-Prefix- List, then another feature Person-Prefix is set to 1.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"For developers of Statistical Machine Translation (SMT) systems, an additional complication is the heterogeneous nature of SMT components (word-alignment model, language model, translation model, etc.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"For all these annotation taks, G\u00c2\u00a8otze developed a series of questions (essentially a decision tree) designed to lead the annotator to the ap propriate judgement.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Evaluation of links A link between two sets is considered correct if the majority of phrases in both sets have the same meaning, i.e. if the link indicates paraphrase.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The result of this is shown in Figure 7.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"However, for multinomial models like our LMs and TMs, there is a one to one correspondence between instances and features, eg the correspondence between a phrase pair (s, t) and its conditional multinomial probability p(s1t).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Previous work on morphological and syntactic disambiguation in Hebrew used different sets of data, different splits, differing annotation schemes, and different evaluation measures.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The types of patterns produced by AutoSlog are outlined in (Riloff, 1996).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Most computational models for coreference resolution rely on properties of the anaphor and candidate antecedent, such as lexical matching, grammatical and syntactic features, semantic agreement, and positional information.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"This group of features attempts to capture such information.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Assigning rhetorical relations thus poses questions that can often be answered only subjectively.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The experimental tests are carried out on the Verbmobil task (GermanEnglish, 8000-word vocabulary), which is a limited-domain spoken-language task.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(We would like to note though that unlike previous boosting algorithms, the CoBoost algorithm presented here is not a boosting algorithm under Valiant's (Valiant 84) Probably Approximately Correct (PAC) model.)","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"If enough parsers suggest that a particular constituent belongs in the parse, we include it.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Also, expanding on the techniques for the automatic generation of extraction patterns (Riloff 96; Sudo 03) using our method, the extraction patterns which have the same meaning can be automatically linked, enabling us to produce the final table fully automatically.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Also, the argument has been made that machine translation performance should be evaluated via task-based evaluation metrics, i.e. how much it assists performing a useful task, such as supporting human translators or aiding the analysis of texts.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Another attempt at using global information can be found in (Borthwick, 1999).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"(b) POS tagging accuracy is lowest for maSdar verbal nouns (VBG,VN) and adjectives (e.g., JJ).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR achieved recall in the 4250% range for both domains, with 76% precision overall for terrorism and 87% precision for natural disasters.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"However, there are phrases which express the same meanings even though they do not share the same keyword.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"By making use of global context, it has achieved excellent results on both MUC6 and MUC7 official test data.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"M(wi) = Li).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"A multicomponent Tree Adjoining Grammar (MCTAG) consists of a finite set of finite elementary tree sets.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"nan2gual 'pumpkin.'","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"lla\/llb and 14a\/14b respectively).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The approach builds from an initial seed set for a category, and is quite similar to the decision list approach described in (Yarowsky 95).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In our model, however, all lattice paths are taken to be a-priori equally likely.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Removing the leaves from the resulting tree yields a parse for L under G, with the desired probabilities.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"A contextual rule considers words surrounding the string in the sentence in which it appears (e.g., a rule that any proper name modified by an appositive whose head is president is a person).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"As observed by Kahane et al. (1998), any (nonprojective) dependency graph can be transformed into a projective one by a lifting operation, which replaces each non-projective arc wj wk by a projective arc wi \u2014* wk such that wi \u2014*\u2217 wj holds in the original graph.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"This representation of structures by substrings and the composition operation by its effect on substrings is related to the work of Rounds (1985).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"8 57.3 +F EA TS be st me dia n 50.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The approach recursively evaluates a quantity Q(C; j), where C is the set of already visited cities and sj is the last visited city.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"If \u00e2\u0080\u009cgun\u00e2\u0080\u009d and \u00e2\u0080\u009crevolver\u00e2\u0080\u009d refer to the same object, then it should also be acceptable to say that Fred was \u00e2\u0080\u009ckilled with a gun\u00e2\u0080\u009d and that the burglar \u00e2\u0080\u009cfireda revolver\u00e2\u0080\u009d.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"In general, several gross corpus statistics favor the ATB, so other factors must contribute to parsing underperformance.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We are unaware of prior results for the Stanford parser.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Each word is terminated by an arc that represents the transduction between f and the part of speech of that word, weighted with an estimated cost for that word.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"3 58.3 40.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"However, it is robust, efficient, and easy to implement.4 To perform the maximization in (7), we used the popular L-BFGS algorithm (Liu and Nocedal, 1989), which requires gradient information.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"If a phrase does not contain any keywords, the phrase is discarded.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Many packages perform language model queries.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Here, annotation proceeds in two phases: first, the domains and the units of IS are marked as such.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"10 Here we use the Good-Turing estimate (Baayen 1989; Church and Gale 1991), whereby the aggregate probability of previously unseen instances of a construction is estimated as ni\/N, where N is the total number of observed tokens and n1 is the number of types observed only once.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"It also incorporates the Good-Turing method (Baayen 1989; Church and Gale 1991) in estimating the likelihoods of previously unseen con\u00c2\u00ad structions, including morphological derivatives and personal names.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"During the search process, a partial hypothesis is extended by choosing a source sentence position, which has not been aligned with a target sentence position yet.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"01 75.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Here we do not submit to this view.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR uses unsupervised learning to acquire this knowledge from plain text without the need for annotated training data.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Nonetheless, the prospect of a network of annotated discourse resources seems particularly promising if not only a single annotation layer is used but a whole variety of them, so that a systematic search for correlations between them becomes possible, which in turn can lead to more explanatory models of discourse structure.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"73 81.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Matsoukas et al (2009) generalize it by learning weights on sentence pairs that are used when estimating relative-frequency phrase-pair probabilities.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"\u00e2\u0080\u0094 similar results have been observed across multiple languages.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"On a set of 11 sentence fragments-the A set-where they reported 100% recall and precision for name identification, we had 73% recall and 80% precision.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This process produces a large set of caseframes coupled with a list of the noun phrases that they extracted.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Only IRSTLM does not support threading.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"1 61.7 37.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The machine learning community has been in a similar situation and has studied the combination of multiple classifiers (Wolpert, 1992; Heath et al., 1996).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We incur some additional memory cost due to storing state in each hypothesis, though this is minimal compared with the size of the model itself.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The DempsterShafer rule for combining pdfs is: to {C}, meaning that it is 70% sure the correct hypothesis is C. The intersection of these sets is the null set because these beliefs are contradictory.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"A final alternate approach would be to combine weighted joint frequencies rather than conditional estimates, ie: cI(s, t) + w,\\(s, t)co(, s, t), suitably normalized.5 Such an approach could be simulated by a MAP-style combination in which separate 0(t) values were maintained for each t. This would make the model more powerful, but at the cost of having to learn to downweight OUT separately for each t, which we suspect would require more training data for reliable performance.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"gaolxing4 'happy' => F.i'JF.i'J Jl!","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"As in boosting, the algorithm works in rounds.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"May 1995).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"6 Our knowledge sources return some sort of probability estimate, although in some cases this estimate is not especially well-principled (e.g., the Recency KS).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"This work was funded by NSF grant IRI-9502312.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We proposed an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The basic word order is VSO, but SVO, VOS, and VO configurations are also possible.2 Nouns and verbs are created by selecting a consonantal root (usually triliteral or quadriliteral), which bears the semantic core, and adding affixes and diacritics.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Recent work by Finkel and Manning (2009) which re-casts Daum\u00b4e\u2019s approach in a hierarchical MAP framework may be applicable to this problem.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"3.4 Salience-based text generation.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We hypothesize that modeling morphological information will greatly constrain the set of possible tags, thereby further refining the representation of the tag lexicon.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(2008) reported agreement between the teams (measured with Evalb) at 93.8% F1, the level of the CTB.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For a given \"word\" in the automatic segmentation, if at least k of the hu\u00c2\u00ad man judges agree that this is a word, then that word is considered to be correct.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"First, it directly encodes linguistic intuitions about POS tag assignments: the model structure reflects the one-tag-per-word property, and a type- level tag prior captures the skew on tag assignments (e.g., there are fewer unique determiners than unique nouns).","label":0},{"source_text":"Here we present two algorithms.","target_text":"One implementation issue deserves some elaboration.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We thank members of the MIT NLP group for their suggestions and comments.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Bikel 75 training trees 5000 10000 15000 Figure 3: Dev set learning curves for sentence lengths \u00e2\u0089\u00a4 70.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"For instance, by altering the emission distribution parameters, Johnson (2007) encourages the model to put most of the probability mass on few tags.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Using cn to denote the number of n-grams, total memory consumption of TRIE, in bits, is plus quantization tables, if used.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This logic applies recursively: if wnf+1 similarly does not extend and has zero log backoff, it too should be omitted, terminating with a possibly empty context.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"However, our full model takes advantage of word features not present in Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Two measures that can be used to compare judgments are: 1.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"7 Conclusion and Future Work.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We use w erations of sampling (see Figure 2 for a depiction).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Semantic expectations are analogous to lexical expectations except that they represent semantic classes rather than nouns.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The cities of the traveling salesman problem correspond to source Table 1: DP algorithm for statistical machine translation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In future work, we plan to follow-up on this approach and investigate other ways that contextual role knowledge can be used.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"In this case we are interested in finding' the maximum probability parse, ri, and Mi is the set of relevant (binary) parsing decisions made by parser i. ri is a parse selected from among the outputs of the individual parsers.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Evaluation Metrics We report three metrics to evaluate tagging performance.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Consequently, we implemented our own annotation tool ConAno in Java (Stede, Heintze 2004), which provides specifically the functionality needed for our purpose.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We use graph-based label propagation for cross-lingual knowledge transfer and use the projected labels as features in an unsupervised model (Berg- Kirkpatrick et al., 2010).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"We show translation results for three approaches: the monotone search (MonS), where no word reordering is allowed (Tillmann, 1997), the quasimonotone search (QmS) as presented in this paper and the IBM style (IbmS) search as described in Section 3.2.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"(f1; ;mg n fl1; l2; l3g ;m) German to English the monotonicity constraint is violated mainly with respect to the German verbgroup.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Linear probing places at most one entry in each bucket.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Non-empty buckets contain an entry belonging to them or to a preceding bucket where a conflict occurred.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Note, however, that there might be situations in which Zco in fact increases.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"This vector tx is constructed for every word in the foreign vocabulary and will be used to provide features for the unsupervised foreign language POS tagger.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"irL as the product of the probability estimate for i\u00c2\u00a5JJ1l., and the probability estimate just derived for unseen plurals in ir,: p(i\u00c2\u00a51J1l.ir,) p(i\u00c2\u00a51J1l.)p(unseen(f,)).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Note that the backoff model assumes that there is a positive correlation between the frequency of a singular noun and its plural.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Second, BABAR performs reliable case resolution to identify anaphora that can be easily resolved using the lexical and syntactic heuristics described in Section 2.1.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Discriminative Instance Weighting for Domain Adaptation in Statistical Machine Translation","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The simplest approach involves scoring the various analyses by costs based on word frequency, and picking the lowest cost path; variants of this approach have been described in Chang, Chen, and Chen (1991) and Chang and Chen (1993).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Unfortunately, there is no standard corpus of Chinese texts, tagged with either single or multiple human judgments, with which one can compare performance of various methods.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"As observed by Kahane et al. (1998), any (nonprojective) dependency graph can be transformed into a projective one by a lifting operation, which replaces each non-projective arc wj wk by a projective arc wi \u2014* wk such that wi \u2014*\u2217 wj holds in the original graph.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While building a machine translation system is a serious undertaking, in future we hope to attract more newcomers to the field by keeping the barrier of entry as low as possible.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Instead, we focused on phrases and set the frequency threshold to 2, and so were able to utilize a lot of phrases while minimizing noise.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In contrast to these approaches, our method directly incorporates these constraints into the structure of the model.","label":0},{"source_text":"There is no global pruning.","target_text":"2.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"It is difficult to motivate these days why one ministry should be exempt from cutbacks \u00e2\u0080\u0094 at the expense of the others.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The P (T |\u00cf\u0088) distribution, in English for instance, should have very low mass for the DT (determiner) tag, since determiners are a very small portion of the vocabulary.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"It is formally straightforward to extend the grammar to include these names, though it does increase the likelihood of overgeneration and we are unaware of any working systems that incorporate this type of name.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"This is equivalent to the assumption used in probability estimation for na\u00efve Bayes classifiers, namely that the attribute values are conditionally independent when the target value is given.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The relevance of the distinction between, say, phonological words and, say, dictionary words is shown by an example like rpftl_A :;!:Hfllil zhong1hua2 ren2min2 gong4he2-guo2 (China people republic) 'People's Republic of China.'","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The types of patterns produced by AutoSlog are outlined in (Riloff, 1996).","label":0},{"source_text":"There is no global pruning.","target_text":"We show translation results for three approaches: the monotone search (MonS), where no word reordering is allowed (Tillmann, 1997), the quasimonotone search (QmS) as presented in this paper and the IBM style (IbmS) search as described in Section 3.2.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We suspect that the higher precision in the disasters domain may be due to its substantially larger training corpus.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The rhetorical structure annotations of PCC have all been converted to URML.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"As you can see in the figure, the accuracy for the domain is quite high except for the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set, which contains various expressions representing different relationships for an IE application.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"This means that the PCC cannot grow particularly quickly.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"To provide a thorough analysis, we evaluated three baselines and two oracles in addition to two variants of our graph-based approach.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Steedman (1986) considers Categorial Grammars in which both the operations of function application and composition may be used, and in which function can specify whether they take their arguments from their right or left.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"For example, in the phrase \u00e2\u0080\u009cCompany-A last week purchased rival Marshalls from Company-B\u00e2\u0080\u009d, the purchased company is Marshalls, not Company-B.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For each domain, we created a semantic dictionary by doing two things.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We use two different similarity functions to define the edge weights among the foreign vertices and between vertices from different languages.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The translation scores for the hypotheses generated with different threshold values t0 are compared to the translation scores obtained with a conservatively large threshold t0 = 10:0 . For each test series, we count the number of sentences whose score is worse than the corresponding score of the test series with the conservatively large threshold t0 = 10:0, and this number is reported as the number of search errors.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The links can solve the problem.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The importance of dictionaries in NERs has been investigated in the literature (Mikheev et al., 1999).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"pre-processing.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In addition, each feature function is a binary function.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We describe a novel approach for inducing unsupervised part-of-speech taggers for languages that have no labeled training data, but have translated text in a resource-rich language.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The simplest approach involves scoring the various analyses by costs based on word frequency, and picking the lowest cost path; variants of this approach have been described in Chang, Chen, and Chen (1991) and Chang and Chen (1993).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Relative pronouns with only 1 NP in scope..","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We have estimated the performance of IdentiFinder ' 99 at 200K words of training data from the graphs.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Clearly this poses a number of research challenges, though, such as the applicability of tag sets across different languages.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Figure 1 reveals that an event that \u00e2\u0080\u009cdamaged\u00e2\u0080\u009d objects may also cause injuries; a disaster that \u00e2\u0080\u009coccurred\u00e2\u0080\u009d may be investigated to find its \u00e2\u0080\u009ccause\u00e2\u0080\u009d; a disaster may \u00e2\u0080\u009cwreak\u00e2\u0080\u009d havoc as it \u00e2\u0080\u009ccrosses\u00e2\u0080\u009d geographic regions; and vehicles that have a \u00e2\u0080\u009cdriver\u00e2\u0080\u009d may also \u00e2\u0080\u009ccarry\u00e2\u0080\u009d items.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"This vector tx is constructed for every word in the foreign vocabulary and will be used to provide features for the unsupervised foreign language POS tagger.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"First, we describe how the caseframes are represented and learned.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Frequency is calculated by counting the number of distinct previous tokens that each token has (e.g., if Electric Corp. is seen 3 times, and Manufacturing Corp. is seen 5 times during training, and Corp. is not seen with any other preceding tokens, then the \u00e2\u0080\u009cfrequency\u00e2\u0080\u009d of Corp. is 2).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Similarly, the tokens and are tested against each list, and if found, a corresponding feature will be set to 1.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This style of naming is never required and seems to be losing currency.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Across eight European languages, our approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The results in Table 1 show PROBING is 81% faster than TRIE, which is in turn 31% faster than the fastest baseline.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"This alters generation of T as follows: n P (T |\u00cf\u0088) = n P (Ti|\u00cf\u0088) i=1 Note that this distribution captures the frequency of a tag across word types, as opposed to tokens.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In the namedentity problem each example is a (spelling,context) pair.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The algorithm in Fig.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"7 Big 5 is the most popular Chinese character coding standard in use in Taiwan and Hong Kong.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"For other languages, we use the CoNLL-X multilingual dependency parsing shared task corpora (Buchholz and Marsi, 2006) which include gold POS tags (used for evaluation).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"markBaseNP indicates these non-recursive nominal phrases.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"A procedural definition to restrict1In the approach described in (Berger et al., 1996), a mor phological analysis is carried out and word morphemes rather than full-form words are used during the search.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Juri Ganitkevitch answered questions about Joshua.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Across all languages, +PRIOR consistently outperforms 1TW, reducing error on average by 9.1% and 5.9% on best and median settings respectively.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For t = 1, T and for j = 1, 2: where 4 = exp(-jg'(xj,i)). practice, this greedy approach almost always results in an overall decrease in the value of Zco.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"52 77.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Case and Zone of and : Similarly, if (or ) is initCaps, a feature (initCaps, zone) (or (initCaps, zone) ) is set to 1, etc. Token Information: This group consists of 10 features based on the string , as listed in Table 1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"com t 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Je voudrais pr\u00b4eciser, a` l\u2019adresse du commissaire Liikanen, qu\u2019il n\u2019est pas ais\u00b4e de recourir aux tribunaux nationaux.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Despite their simplicity, uni- gram weights have been shown as an effective feature in segmentation models (Dyer, 2009).13 The joint parser\/segmenter is compared to a pipeline that uses MADA (v3.0), a state-of-the-art Arabic segmenter, configured to replicate ATB segmentation (Habash and Rambow, 2005).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This representation gives ir, an appropriate morphological decomposition, pre\u00c2\u00ad serving information that would be lost by simply listing ir, as an unanalyzed form.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The Recency KS computes the distance between the candidate and the anaphor relative to its scope.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Because we are working with only three parsers, the only situation in which context will help us is when it can indicate we should choose to believe a single parser that disagrees with the majority hypothesis instead of the majority hypothesis itself.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The rationale for treating these semantic labels differently is that they are specific and reliable (as opposed to the WordNet classes, which are more coarse and more noisy due to polysemy).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"\u00e2\u0088\u0097NP NP PP R) and \u00e2\u0088\u0097NP NP ADJP R) are both iDafa attachment.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"where the husband's family name is optionally prepended to the woman's full name; thus ;f:*lf#i xu3lin2-yan2hai3 would represent the name that Ms. Lin Yanhai would take if she married someone named Xu.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In this paper, we make a simplifying assumption of one-tag-per-word.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Parser 3, the most accurate parser, was chosen 71% of the time, and Parser 1, the least accurate parser was chosen 16% of the time.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In Input: {(x1,i, Initialize: Vi, j : e(xi) = 0.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The anaphor and antecedent appear in boldface.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Relative pronouns with only 1 NP in scope..","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, co-occurring caseframes may reflect synonymy (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> abducted\u00e2\u0080\u009d) or related events (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> released\u00e2\u0080\u009d).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Specifically, for the ith word type, the set of token-level tags associated with token occurrences of this word, denoted t(i), must all take the value Ti to have nonzero mass. Thus in the context of Gibbs sampling, if we want to block sample Ti with t(i), we only need sample values for Ti and consider this setting of t(i).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Each xii is a member of X, where X is a set of possible features.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Without using the same test corpus, direct comparison is obviously difficult; fortunately, Chang et al. include a list of about 60 sentence fragments that exemplify various categories of performance for their system.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In addition to the named-entity string (Maury Cooper or Georgia), a contextual predictor was also extracted.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Standard SMT systems have a hierarchical parameter structure: top-level log-linear weights are used to combine a small set of complex features, interpreted as log probabilities, many of which have their own internal parameters and objectives.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Unlabeled examples in the named-entity classification problem can reduce the need for supervision to a handful of seed rules.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"They are set to fixed constants.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"72 77.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The segmenter handles the grouping of hanzi into words and outputs word pronunciations, with default pronunciations for hanzi it cannot group; we focus here primarily on the system's ability to segment text appropriately (rather than on its pronunciation abilities).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"orthographic words are thus only a starting point for further analysis and can only be regarded as a useful hint at the desired division of the sentence into words.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Of course, we are primarily interested in applying our techniques to languages for which no labeled resources are available.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Graph construction for structured prediction problems such as POS tagging is non-trivial: on the one hand, using individual words as the vertices throws away the context necessary for disambiguation; on the other hand, it is unclear how to define (sequence) similarity if the vertices correspond to entire sentences.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Similarly Figures 1 and 2 show how the isolated constituent precision varies by sentence length and the size of the span of the hypothesized constituent.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We offer a state function s(wn1) = wn\ufffd where substring wn\ufffd is guaranteed to extend (to the right) in the same way that wn1 does for purposes of language modeling.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The search starts in hypothesis (f;g; 0) and ends in the hypotheses (f1; ; Jg; j), with j 2 f1; ; Jg.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"It's not clear how to apply these methods in the unsupervised case, as they required cross-validation techniques: for this reason we use the simpler smoothing method shown here. input to the unsupervised algorithm is an initial, &quot;seed&quot; set of rules.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"5.1 Data Sets.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"A procedural definition to restrict1In the approach described in (Berger et al., 1996), a mor phological analysis is carried out and word morphemes rather than full-form words are used during the search.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The resulting structural differences between tree- banks can account for relative differences in parsing performance.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(5) and ht into Equ.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"This problem arises because our keywords consist of only one word.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Having explained the various layers of annotation in PCC, we now turn to the question what all this might be good for.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"tai2du2 'Taiwan Independence.'","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Unsupervised learning approaches appear to be a natural solution to this problem, as they require only unannotated text for training models.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"For each set, the phrases with bracketed frequencies are considered not paraphrases in the set.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(3)), with one term for each classifier.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The following recursive equation is evaluated: Qe0 (e; S; C; j) = (2) = p(fj je) max \u00c3\u0086;e00 np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) max (S0;j0) (S0 ;Cnfjg;j0)!(S;C;j) j02Cnfjg Qe00 (e0; S0;C n fjg; j0)o: The search ends in the hypotheses (I; f1; ; Jg; j).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Experiments using data from the Prague Dependency Treebank show that the combined system can handle nonprojective constructions with a precision sufficient to yield a significant improvement in overall parsing accuracy.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"All the sentences have been analyzed by our chunker and NE tag- ger.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We show that noun-noun vs. discourse-level coordination ambiguity in Arabic is a significant source of parsing errors (Table 8c).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"However, the learning curves in Figure 3 show that the Berkeley parser does not exceed our manual grammar by as wide a margin as has been shown for other languages (Petrov, 2009).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Month Names, Days of the Week, and Numbers: If is initCaps and is one of January, February, . . .","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"At this step, we will try to link those sets, and put them into a single cluster.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"But in most cases they can be used interchangably.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The following auxiliary quantity is defined: Qe0 (e; C; j) := probability of the best partial hypothesis (ei 1; bi 1), where C = fbkjk = 1; ; ig, bi = j, ei = e and ei\u00f4\u0080\u0080\u00801 = e0.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"As seen by the drop in average individual parser performance baseline, the introduced parser does not perform very well.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Note that these observa sider suffix features, capitalization features, punctuation, and digit features.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"In this paper we have proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MUC7 has also seen hybrids of statistical NERs and hand-coded systems (Mikheev et al., 1998; Borthwick, 1999), notably Mikheev' s system, which achieved the best performance of 93.39% on the official NE test data.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"computing the recall of the other's judgments relative to this standard.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The accuracy of the sets in representing paraphrase ranged from 73% to 99%, depending on the NE categories and set sizes; the accuracy of the links for two evaluated domains was 73% and 86%.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"58 95.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Recently, this topic has been getting more attention, as is evident from the Paraphrase Workshops in 2003 and 2004, driven by the needs of various NLP applications.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We would like to thank Ryan McDonald for numerous discussions on this topic.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Arabic is a morphologically rich language with a root-and-pattern system similar to other Semitic languages.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"If there is a frequent multi-word sequence in a domain, we could use it as a keyword candidate.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Section 2 describes our baseline techniques for SMT adaptation, and section 3 describes the instance-weighting approach.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We therefore used the arithmetic mean of each interjudge precision-recall pair as a single measure of interjudge similarity.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In the second part of the experiment, we applied the inverse transformation based on breadth-first search under the three different encoding schemes.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"\u2022 Similarly, when the na\u00efve Bayes classifier is configured such that the constituents require estimated probabilities strictly larger than 0.5 to be accepted, there is not enough probability mass remaining on crossing brackets for them to be included in the hypothesis.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Saving state allows our code to walk the data structure exactly once per query.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Finally, we note that Jiang\u2019s instance-weighting framework is broader than we have presented above, encompassing among other possibilities the use of unlabelled IN data, which is applicable to SMT settings where source-only IN corpora are available.","label":0},{"source_text":"There is no global pruning.","target_text":"For a given partial hypothesis (C; j), the order in which the cities in C have been visited can be ignored (except j), only the score for the best path reaching j has to be stored.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We trained this model by optimizing the following objective function: Note that this involves marginalizing out all possible state configurations z for a sentence x, resulting in a non-convex objective.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, management succession systems must distinguish between a person who is fired and a person who is hired.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"na me =>1 ha nzi fa mi ly 2 ha nzi gi ve n 3.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"To show that the derivation trees of any grammar in LCFRS is a local set, we can rewrite the annotated derivation trees such that every node is labelled by a pair to include the composition operations.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We are not claiming that this method is almighty.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"from the subset of the United Informatics corpus not used in the training of the models.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"markBaseNP indicates these non-recursive nominal phrases.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"For instance, the sentence Similar improvements in haemoglobin levels were reported in the scientific literature for other epoetins would likely be considered domain-specific despite the presence of general phrases like were reported in.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"hanzi in the various name positions, derived from a million names.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The second algorithm builds on a boosting algorithm called AdaBoost (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"For monolingual treebank data we relied on the CoNLL-X and CoNLL-2007 shared tasks on dependency parsing (Buchholz and Marsi, 2006; Nivre et al., 2007).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"37 79.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Mutual information was shown to be useful in the segmentation task given that one does not have a dictionary.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Compared to decoding, this task is cache-unfriendly in that repeated queries happen only as they naturally occur in text.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Again, the idea is that having a picture of syntax, co-reference, and sentence-internal information structure at one\u00e2\u0080\u0099s disposal should aid in finding models of discourse structure that are more explanatory and can be empirically supported.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Section 4 presents experimen tal results on two corpora: the MUC4 terrorism corpus, and Reuters texts about natural disasters.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"ogy (Koskenniemi 1983; Antworth 1990; Tzoukermann and Liberman 1990; Karttunen, Kaplan, and Zaenen 1992; Sproat 1992); we represent the fact that ir, attaches to nouns by allowing t:-transitions from the final states of all noun entries, to the initial state of the sub-WFST representing f,.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Each xt E 2x is the set of features constituting the ith example.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Contains digits.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This flexibility, along with the simplicity of implementation and expansion, makes this framework an attractive base for continued research.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"One of the difficulties in Natural Language Processing is the fact that there are many ways to express the same thing or event.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The DL-CoTrain algorithm can be motivated as being a greedy method of satisfying the above 2 constraints.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"One possibility is to use n-grams based on mutual information.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We start with noun features since written Arabic contains a very high proportion of NPs.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The unlabeled data gives many such &quot;hints&quot; that two features should predict the same label, and these hints turn out to be surprisingly useful when building a classifier.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"As we shall argue, the semantic class affiliation of a hanzi constitutes useful information in predicting its properties.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We used the MUC4 terrorism corpus (MUC4 Proceedings, 1992) and news articles from the Reuter\u00e2\u0080\u0099s text collection8 that had a subject code corresponding to natural disasters.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"of Articles No.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Out-of-Vocabulary: We derived a lexicon list from WordNet 1.6, and words that are not found in this list have a feature out-of-vocabulary set to 1.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"By making use of global context, it has achieved excellent results on both MUC6 and MUC7 official test data.","label":1},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Each trie node contains a sorted array of entries and they use binary search.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"64 94.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We would like to thank Prof. Ralph Grish- man, Mr. Takaaki Hasegawa and Mr. Yusuke Shinyama for useful comments, discussion and evaluation.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We can now compare this algorithm to that of (Yarowsky 95).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"However, for multinomial models like our LMs and TMs, there is a one to one correspondence between instances and features, eg the correspondence between a phrase pair (s, t) and its conditional multinomial probability p(s1t).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"An additional case of super-segmental morphology is the case of Pronominal Clitics.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The confidence intervals are computed by bootstrap resampling for BLEU, and by standard significance testing for the manual scores, as described earlier in the paper.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"JA DE G O L D G R AS S SI C K NE SS DE AT H R A T 14.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The system of Berg-Kirkpatrick et al.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"CFG's, TAG's, MCTAG's and HG's are all members of this class since they satisfy these restrictions.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"It is generally considered to be fast (Pauls 29 \u2212 1 probabilities and 2' \u2212 2 non-zero backoffs. and Klein, 2011), with a default implementation based on hash tables within each trie node.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"t(i).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We have argued that the proposed method performs well.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The tokens w are generated by token-level tags t from an HMM parameterized by the lexicon structure.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We are especially grateful to Taylor Berg- Kirkpatrick for running additional experiments.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Figure 1: Translation of PCC sample commentary (STTS)2.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The scoping heuristics are based on the anaphor type: for reflexive pronouns the scope is the current clause, for relative pronouns it is the prior clause following its VP, for personal pronouns it is the anaphor\u00e2\u0080\u0099s sentence and two preceding sentences, and for definite NPs it is the anaphor\u00e2\u0080\u0099s sentence and eight preceding sentences.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Step 1.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Adding the isolated constituents to our hypothesis parse could increase our expected recall, but in the cases we investigated it would invariably hurt our precision more than we would gain on recall.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Quite often, though, these directives fulfill the goal of increasing annotator agreement without in fact settling the theoretical question; i.e., the directives are clear but not always very well motivated.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"On average, 6 reference translations per automatic translation are available.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We extend the Stanford parser to accept pre-generated lattices, where each word is represented as a finite state automaton.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Tag set As is standard, for all experiments, we set the number of latent model tag states to the size of the annotated tag set.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The human evaluators were a non-native, fluent Arabic speaker (the first author) for the ATB and a native English speaker for the WSJ.7 Table 5 shows type- and token-level error rates for each corpus.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"However, TRIE partitions storage by n-gram length, so walking the trie reads N disjoint pages.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The increase is generally higher for PDT than for DDT, which indicates a greater diversity in non-projective constructions.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Sentence pairs are the natural instances for SMT, but sentences often contain a mix of domain-specific and general language.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"It is difficult to directly compare the Matsoukas et al results with ours, since our out-of-domain corpus is homogeneous; given heterogeneous training data, however, it would be trivial to include Matsoukas-style identity features in our instance-weighting model.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Unfortunately, modifying the model to account for these kind of dependencies is not at all straightforward.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"1.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Mi(c) is a binary function returning t when parser i (from among the k parsers) suggests constituent c should be in the parse.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"In our particular case the majority requires the agreement of only two parsers because we have only three.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Table 4 shows translation results for the three approaches.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We have not yet tried this.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The TRIE model continues to use the least memory of ing (-P) with MAP POPULATE, the default.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The system of Berg-Kirkpatrick et al.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Similar advances have been made in machine translation (Frederking and Nirenburg, 1994), speech recognition (Fiscus, 1997) and named entity recognition (Borthwick et al., 1998).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, both a chair and a truck would be labeled as artifacts, but this does not at all suggest that they are coreferent.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The errors shown are from the Berkeley parser output, but they are representative of the other two parsing models.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The breakdown of the different types of words found by ST in the test corpus is given in Table 3.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"These systems are similar to those described by Pollard (1984) as Generalized Context-Free Grammars (GCFG's).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Note that because we extracted only high-confidence alignments, many foreign vertices will not be connected to any English vertices.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"This paper does not necessarily reflect the position of the U.S. Government.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"This group of features attempts to capture such information.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Because the information flow in our graph is asymmetric (from English to the foreign language), we use different types of vertices for each language.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Although embedding this version of LCFRS's in the framework of ILFP developed by Rounds (1985) is straightforward, our motivation was to capture properties shared by a family of grammatical systems and generalize them defining a class of related formalisms.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We applied the AutoSlog system (Riloff, 1996) to our unannotated training texts to generate a set of extraction patterns for each domain.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, they list two sets, one consisting of 28 fragments and the other of 22 fragments, in which they had 0% recall and precision.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Instead, the designs of the various annotation layers and the actual annotation work are results of a series of diploma theses, of students\u00e2\u0080\u0099 work in course projects, and to some extent of paid assistentships.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The breakdown of the different types of words found by ST in the test corpus is given in Table 3.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For corporate suffixes, a list of tokens cslist that occur frequently as the last token of an organization name is collected from the training data.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(3)), with one term for each classifier.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In the case of, the most common usage is as an adverb with the pronunciation jiangl, so that variant is assigned the estimated cost of 5.98, and a high cost is assigned to nominal usage with the pronunciation jiang4.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"\u00e2\u0080\u009d).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"F1 85 Berkeley 80 Stanford.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"7 Big 5 is the most popular Chinese character coding standard in use in Taiwan and Hong Kong.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Using this heuristic, BABAR identifies existential definite NPs in the training corpus using our previous learning algorithm (Bean and Riloff, 1999) and resolves all occurrences of the same existential NP with each another.1 2.1.2 Syntactic Seeding BABAR also uses syntactic heuristics to identify anaphors and antecedents that can be easily resolved.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The similarities become apparent when they are studied at the level of derivation structures: derivation nee sets of CFG's, HG's, TAG's, and MCTAG's are all local sets.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The simplest version of the maximum matching algorithm effectively deals with ambiguity by ignoring it, since the method is guaranteed to produce only one segmentation.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"This feature ft incorporates information from the smoothed graph and prunes hidden states that are inconsistent with the thresholded vector tx.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Indeed there are several open issues.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"This design does not guarantee \u00e2\u0080\u009cstructural zeros,\u00e2\u0080\u009d but biases towards sparsity.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"encodes the one tag per word constraint and is uni form over type-level tag assignments.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"A generative model was applied (similar to naive Bayes) with the three labels as hidden vanables on unlabeled examples, and observed variables on (seed) labeled examples.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The key point is that the second constraint can be remarkably powerful in reducing the complexity of the learning problem.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Previous reports on Chinese segmentation have invariably cited performance either in terms of a single percent-correct score, or else a single precision-recall pair.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The fraction of buckets that are empty is m\u22121 m , so average lookup time is O( m 1) and, crucially, constant in the number of entries.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Hyperparameter settings are sorted according to the median one-to-one metric over runs.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"There has been additional recent work on inducing lexicons or other knowledge sources from large corpora.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Without using the same test corpus, direct comparison is obviously difficult; fortunately, Chang et al. include a list of about 60 sentence fragments that exemplify various categories of performance for their system.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"To measure the contribution of each modification, a third, intermediate algorithm, Yarowsky-cautious was also tested.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We are especially grateful to Taylor Berg- Kirkpatrick for running additional experiments.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Figure 3 Figure 1.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Evaluation Metrics We report three metrics to evaluate tagging performance.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Put another way, the minimum of Equ.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Besides information structure, the second main goal is to enhance current models of rhetorical structure.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"4.5 Transliterations of Foreign Words.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Here we propose a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"0 X u} \"' o; .2 X X><X X XX X X X X X X x X X X X X x X V X X X X .;t'*- XXX:OX X X X X X X 9 x X X XX XX X X X X X X X XXX:< X X>O<XX>!KXX XI<>< \u00c2\u00bbC X X XX :X: X X \"' X X XX >OO<X>D<XIK X X X X X X --XX\u00c2\u00bb: XXX X X\u00c2\u00bbC X X\u00c2\u00abX...C:XXX X Xll< X X ><XX>IIC:liiC:oiiiiCI--8!X:liiOC!I!S8K X X X 10 100 1000 10000 log(F)_base: R\"2=0.20 (p < 0.005) X 100000 Figure 6 Plot of log frequency of base noun, against log frequency of plural nouns.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The corpus-based statistical parsing community has many fast and accurate automated parsing systems, including systems produced by Collins (1997), Charniak (1997) and Ratnaparkhi (1997).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"of Tokens MENERGI 318 160,000 200 180,000 IdentiFinder \u00e2\u0080\u0093 650,000 \u00e2\u0080\u0093 790,000 MENE \u00e2\u0080\u0093 \u00e2\u0080\u0093 350 321,000 Table 4: Training Data MUC7 test accuracy.2 For MUC6, the reduction in error due to global features is 27%, and for MUC7,14%.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We used a standard one-pass phrase-based system (Koehn et al., 2003), with the following features: relative-frequency TM probabilities in both directions; a 4-gram LM with Kneser-Ney smoothing; word-displacement distortion model; and word count.","label":0},{"source_text":"There is no global pruning.","target_text":"In German, the verbgroup usually consists of a left and a right verbal brace, whereas in English the words of the verbgroup usually form a sequence of consecutive words.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If all three sentences are in the same document, then even a human will find it difficult to classify McCann in (1) into either person or organization, unless there is some other information provided.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Depending on the threshold t0, the search algorithm may miss the globally optimal path which typically results in additional translation errors.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Given a sufficient number of randomly drawn unlabeled examples (i.e., edges), we will induce two completely connected components that together span the entire graph.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We tagged each noun with the top-level semantic classes assigned to it in Word- Net.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The natural language processing community is in the strong position of having many available approaches to solving some of its most fundamental problems.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The computation of Pfr1(c)1Mi M k (C)) has been sketched before in Equations 1 through 4.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Naseem et al. (2009) and Snyder et al.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"1 is given in Fig.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Storing state therefore becomes a time-space tradeoff; for example, we store state with partial hypotheses in Moses but not with each phrase.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We used it to score all phrase pairs in the OUT table, in order to provide a feature for the instance-weighting model.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We adopted the MUC6 guidelines for evaluating coreference relationships based on transitivity in anaphoric chains.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Such analyzers propose multiple segmentation possibilities and their corresponding analyses for a token in isolation but have no means to determine the most likely ones.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Previous approaches have tried to find examples that are similar to the target domain.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The distribution specifies the relative weight, or importance, of each example \u2014 typically, the weak learner will attempt to minimize the weighted error on the training set, where the distribution specifies the weights.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In our second model GTvpi we also distinguished finite and non-finite verbs and VPs as 10Lattice parsing can be performed by special initialization of the chart in a CKY parser (Chappelier et al., 1999).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Text generation, or at least the two phases of text planning and sentence planning, is a process driven partly by well-motivated choices (e.g., use this lexeme X rather than that more colloquial near-synonym Y ) and partly by con tation like that of PCC can be exploited to look for correlations in particular between syntactic structure, choice of referring expressions, and sentence-internal information structure.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We propose an unsupervised method to discover paraphrases from a large untagged corpus, without requiring any seed phrase or other cue.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Denote by g3(x) = Et crithl(x) , j E {1,2} the unthresholded strong-hypothesis (i.e., f3 (x) = sign(gi (x))).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In Eq.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Because of this threshold, very few NE instance pairs could be used and hence the variety of phrases was also limited.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"47 78.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"able at http:\/\/nlp.stanford.edu\/projects\/arabic.shtml.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The Expectation Maximization (EM) algorithm (Dempster, Laird and Rubin 77) is a common approach for unsupervised training; in this section we describe its application to the named entity problem.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In addition to the named-entity string (Maury Cooper or Georgia), a contextual predictor was also extracted.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"It is also true of the adaptation of the Collins parser for Czech (Collins et al., 1999) and the finite-state dependency parser for Turkish by Oflazer (2003).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"What is important and is not expressed by the notation is the so-called coverage constraint: each source position j should be 'hit' exactly once by the path of the inverted alignment bI 1 = b1:::bi:::bI . Using the inverted alignments in the maximum approximation, we obtain as search criterion: max I (p(JjI) max eI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) max bI 1 I Yi=1 [p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])) = = max I (p(JjI) max eI 1;bI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])); where the two products over i have been merged into a single product over i. p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) is the trigram language model probability.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The breakdown of the different types of words found by ST in the test corpus is given in Table 3.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Wu and Fung introduce an evaluation method they call nk-blind.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"1 2 3.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Step 4.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This is appropriate in cases where it is sanctioned by Bayes\u2019 law, such as multiplying LM and TM probabilities, but for adaptation a more suitable framework is often a mixture model in which each event may be generated from some domain.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The use of the Good-Turing equation presumes suitable estimates of the unknown expectations it requires.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Mixing, smoothing, and instance-feature weights are learned at the same time using an efficient maximum-likelihood procedure that relies on only a small in-domain development corpus.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Taking only the highest frequency rules is much &quot;safer&quot;, as they tend to be very accurate.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Month Names, Days of the Week, and Numbers: If is initCaps and is one of January, February, . . .","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Nonetheless there is no alternative to Reiche\u00e2\u0080\u0099s plan.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For novel texts, no lexicon that consists simply of a list of word entries will ever be entirely satisfactory, since the list will inevitably omit many constructions that should be considered words.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Others depend upon various lexical heuris\u00c2\u00ad tics: for example Chen and Liu (1992) attempt to balance the length of words in a three-word window, favoring segmentations that give approximately equal length for each word.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"However, there is a crucial difference: the morphological probabilities in their model come from discriminative models based on linear context.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Each trie node is individually allocated and full 64-bit pointers are used to find them, wasting memory.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The domains are the linguistic spans that are to receive an IS-partitioning, and the units are the (smaller) spans that can play a role as a constituent of such a partitioning.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"In focus is in particular the correlation with rhetorical structure, i.e., the question whether specific rhetorical relations \u00e2\u0080\u0094 or groups of relations in particular configurations \u00e2\u0080\u0094 are signalled by speakers with prosodic means.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"RandLM and SRILM also remove context that will not extend, but SRILM performs a second lookup in its trie whereas our approach has minimal additional cost.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The evidence may not say anything about whether A is more likely than B, only that C and D are not likely.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Once the lexicon has been drawn, the model proceeds similarly to the standard token-level HMM: Emission parameters \u00ce\u00b8 are generated conditioned on tag assignments T . We also draw transition parameters \u00cf\u0086.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"4 53.7 43.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"After adding a ROOT node to all trees, we train a grammar using six split-and- merge cycles and no Markovization.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Second, we identified the 100 most frequent nouns in the training corpus and manually labeled them with semantic tags.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models (e.g., (Grosz et al., 1995; Grosz and Sidner, 1998)), syntactic algorithms (e.g., (Hobbs, 1978; Lappin and Le- ass, 1994)), and supervised machine learning systems (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The observed performance gains, coupled with the simplicity of model implementation, makes it a compelling alternative to existing more complex counterparts.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The Bayes models were able to achieve significantly higher precision than their non-parametric counterparts.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The current work treats both segmental and super-segmental phenomena, yet we note that there may be more adequate ways to treat supersegmental phenomena assuming Word-Based morphology as we explore in (Tsarfaty and Goldberg, 2008).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"From this we see that a finer-grained model for parser combination, at least for the features we have examined, will not give us any additional power.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This is not unreasonable given the application to phrase pairs from OUT, but it suggests that an interesting alternative might be to use a plain log-linear weighting function exp(Ei Aifi(s, t)), with outputs in [0, oo].","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In many cases, inspection of either the spelling or context alone is sufficient to classify an example.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"For rules p : A fpo such that fp is constant function, giving an elementary structure, fp is defined such that fp() = (Si ... xi() where each z is a constant string.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"See Table 2 for the tag set size of other languages.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Tables 4 and 5 show BABAR\u00e2\u0080\u0099s performance when just one contextual role knowledge source is used at a time.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Thus it is possible, for illustration, to look for a noun phrase (syntax tier) marked as topic (information structure tier) that is in a bridging relation (co-reference tier) to some other noun phrase.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This technique was introduced by Clarkson and Rosenfeld (1997) and is also implemented by IRSTLM and BerkeleyLM\u2019s compressed option.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Fortunately, there are only a few hundred hanzi that are particularly common in transliterations; indeed, the commonest ones, such as E. bal, m er3, and iij al are often clear indicators that a sequence of hanzi containing them is foreign: even a name like !:i*m xia4mi3-er3 'Shamir,' which is a legal Chi\u00c2\u00ad nese personal name, retains a foreign flavor because of liM.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"and f,.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Every token is independent of the others, and the sentence lattice is in fact a concatenation of smaller lattices, one for each token.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The NP is a complement to a preposition, which is the head of a PP.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We include a list of per-category results for selected phrasal labels, POS tags, and dependencies in Table 8.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Statistical methods seem particularly applicable to the problem of unknown-word identification, especially for constructions like names, where the linguistic constraints are minimal, and where one therefore wants to know not only that a particular se\u00c2\u00ad quence of hanzi might be a name, but that it is likely to be a name with some probabil\u00c2\u00ad ity.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Previous approaches have tried to find examples that are similar to the target domain.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Assume that the two classifiers are &quot;rote learners&quot;: that is, 1.1 and 12 are defined through look-up tables that list a label for each member of X1 or X2.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Moses keeps language models and many other resources in static variables, so these are still resident in memory.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Finally, we thank Kuzman Ganchev and the three anonymous reviewers for helpful suggestions and comments on earlier drafts of this paper.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Every parse \u03c0 selects a specific morphological segmentation (l1...lk) (a path through the lattice).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The types of patterns produced by AutoSlog are outlined in (Riloff, 1996).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"For one thing, it is not clear who is to receive settlements or what should happen in case not enough teachers accept the offer of early retirement.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This would result in better rest cost estimation and better pruning.10 In general, tighter, but well factored, integration between the decoder and language model should produce a significant speed improvement.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Like HG's, TAG's, and MCTAG's, members of LCFRS can manipulate structures more complex than terminal strings and use composition operations that are more complex that concatenation.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"But for eign learners are often surprised by the verbless predications that are frequently used in Arabic.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We have checked if there are similar verbs in other major domains, but this was the only one.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"PoS tags impose a unique morphological segmentation on surface tokens and present a unique valid yield for syntactic trees.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The BerkeleyLM direct-mapped cache is in principle faster than caches implemented by RandLM and by IRSTLM, so we may write a C++ equivalent implementation as future work.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Applications such as machine translation use language model probability as a feature to assist in choosing between hypotheses.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The wounds are still healing.), entity-attribute (e.g., She 2001), who determined that in their corpus of German computer tests, 38% of relations were lexically signalled.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This style of naming is never required and seems to be losing currency.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In any event, to date, we have not compared different methods for deriving the set of initial frequency estimates.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"6) are noisy, the results confirm that label propagation within the foreign language part of the graph adds significant quality for every language.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"3 54.4 33.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"If all three sentences are in the same document, then even a human will find it difficult to classify McCann in (1) into either person or organization, unless there is some other information provided.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Given n-gram counts {cn}Nn=1, we use Flog2 c1] bits per vocabulary identifier and Flog2 cn] per index into the table of ngrams.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The PROBING model was designed to improve upon SRILM by using linear probing hash tables (though not arranged in a trie), allocating memory all at once (eliminating the need for full pointers), and being easy to compile.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"In both cases the investigators were able to achieve significant improvements over the previous best tagging results.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Our work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"It is difficult when IN and OUT are dissimilar, as they are in the cases we study.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"In this case we are interested in finding' the maximum probability parse, ri, and Mi is the set of relevant (binary) parsing decisions made by parser i. ri is a parse selected from among the outputs of the individual parsers.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Since the parameter and token components will remain fixed throughout experiments, we briefly describe each.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"There are two weaknesses in Chang et al.'s model, which we improve upon.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Therefore, a populated probing hash table consists of an array of buckets that contain either one entry or are empty.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The weak learner for two-class problems computes a weak hypothesis h from the input space into the reals (h : 2x -4 R), where the sign4 of h(x) is interpreted as the predicted label and the magnitude I h(x)I is the confidence in the prediction: large numbers for I h(x)I indicate high confidence in the prediction, and numbers close to zero indicate low confidence.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, if X and Y are coreferent, then both X and Y are considered to co-occur with the caseframe that extracts X as well as the caseframe that extracts Y. We will refer to the set of nouns that co-occur with a caseframe as the lexical expectations of the case- frame.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Unlabeled examples in the named-entity classification problem can reduce the need for supervision to a handful of seed rules.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, some caveats are in order in comparing this method (or any method) with other approaches to seg\u00c2\u00ad mentation reported in the literature.","label":0},{"source_text":"There is no global pruning.","target_text":"4.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"2.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Tokens were converted to vocabulary identifiers in advance and state was carried from each query to the next.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Segments with the same surface form but different PoS tags are treated as different lexemes, and are represented as separate arcs (e.g. the two arcs labeled neim from node 6 to 7).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"During the search process, a partial hypothesis is extended by choosing a source sentence position, which has not been aligned with a target sentence position yet.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"At the same time, the n-gram error rate is sensitive to samples with extreme n-gram counts.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We have described two data structures for language modeling that achieve substantial reductions in time and memory cost.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"When the parser is trained on the transformed data, it will ideally learn not only to construct projective dependency structures but also to assign arc labels that encode information about lifts.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"where the husband's family name is optionally prepended to the woman's full name; thus ;f:*lf#i xu3lin2-yan2hai3 would represent the name that Ms. Lin Yanhai would take if she married someone named Xu.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As with personal names, we also derive an estimate from text of the probability of finding a transliterated name of any kind (PTN).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"vierten 12.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This is not unreasonable given the application to phrase pairs from OUT, but it suggests that an interesting alternative might be to use a plain log-linear weighting function exp(Ei Aifi(s, t)), with outputs in [0, oo].","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Finally, the focus\/background partition is annotated, together with the focus question that elicits the corresponding answer.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The precision and recall measures (described in more detail in Section 3) used in evaluating Treebank parsing treat each constituent as a separate entity, a minimal unit of correctness.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The DempsterShafer rule for combining pdfs is: to {C}, meaning that it is 70% sure the correct hypothesis is C. The intersection of these sets is the null set because these beliefs are contradictory.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The returned state s(wn1) may then be used in a followon query p(wn+1js(wn1)) that extends the previous query by one word.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"A simple lexicalized PCFG with second order Markovization gives relatively poor performance: 75.95% F1 on the test set.8 But this figure is surprisingly competitive with a recent state-of-the-art baseline (Table 7).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For example Chen and Liu (1992) report precision and recall rates of over 99%, but this counts only the words that occur in the test corpus that also occur in their dictionary.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"On each language we investigate the contribution of each component of our model.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The number of top-ranked pairs to retain is chosen to optimize dev-set BLEU score.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"For graph propagation, the hyperparameter v was set to 2 x 10\u22126 and was not tuned.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This leads to word- and constituent-boundaries discrepancy, which breaks the assumptions underlying current state-of-the-art statistical parsers.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"One of such approaches uses comparable documents, which are sets of documents whose content are found\/known to be almost the same, such as different newspaper stories about the same event [Shinyama and Sekine 03] or different translations of the same story [Barzilay 01].","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The importance of dictionaries in NERs has been investigated in the literature (Mikheev et al., 1999).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The entire treebank is used in the experiment, but only primary dependencies are considered.4 In all experiments, punctuation tokens are included in the data but omitted in evaluation scores.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We use the default inference parameters.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"At present, the \u00e2\u0080\u0098Potsdam Commentary Corpus\u00e2\u0080\u0099 (henceforth \u00e2\u0080\u0098PCC\u00e2\u0080\u0099 for short) consists of 170 commentaries from Ma\u00c2\u00a8rkische Allgemeine Zeitung, a German regional daily.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We also add an annotation for one-level iDafa (oneLevelIdafa) constructs since they make up more than 75% of the iDafa NPs in the ATB (Gabbard and Kulick, 2008).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Other errors include NE tagging errors and errors due to a phrase which includes other NEs.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Berland and Charniak 99) describe a method for extracting parts of objects from wholes (e.g., &quot;speedometer&quot; from &quot;car&quot;) from a large corpus using hand-crafted patterns.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"A similar maximumlikelihood approach was used by Foster and Kuhn (2007), but for language models only.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Since the parameter and token components will remain fixed throughout experiments, we briefly describe each.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"(Hartmann 1984), for example, used the term Reliefgebung to characterize the distibution of main and minor information in texts (similar to the notion of nuclearity in RST).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The horizontal axis in this plot represents the most significant dimension, which explains 62% of the variation.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thus, the derivation trees for TAG's have the same structure as local sets.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Further, the special hash 0 suffices to flag empty buckets.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Trees are composed using an operation called adjoining, which is defined as follows.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"3.3 Evaluation Results.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"This remains as future work.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Suppose -y results from the adjunction of 71, ,-y, at the k distinct tree addresses 141, , nk in some elementary tree 7', respectively.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The toplevel weights are trained to maximize a metric such as BLEU on a small development set of approximately 1000 sentence pairs.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"To evaluate proper-name identification, we randomly se\u00c2\u00ad lected 186 sentences containing 12,000 hanzi from our test corpus and segmented the text automatically, tagging personal names; note that for names, there is always a sin\u00c2\u00ad gle unambiguous answer, unlike the more general question of which segmentation is correct.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We present KenLM, a library that implements two data structures for efficient language model queries, reducing both time and costs.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Examples will usually be accompanied by a translation, plus a morpheme-by-morpheme gloss given in parentheses whenever the translation does not adequately serve this purpose.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Table 3 contains the results for evaluating our systems on the test set (section 22).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"In other words, the set of hidden states F was chosen to be the fine set of treebank tags.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The next two rows are results of oracle experiments.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The question is how to normalize the probabilities in such a way that smaller groupings have a better shot at winning.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"To establish a soft correspondence between the two languages, we use a second similarity function, which leverages standard unsupervised word alignment statistics (\u00a73.3).3 Since we have no labeled foreign data, our goal is to project syntactic information from the English side to the foreign side.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The trie data structure is commonly used for language modeling.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"4 65.9 48.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Unlike the WSJ corpus which has a high frequency of rules like VP \u00e2\u0086\u0092VB PP, Arabic verb phrases usually have lexi calized intervening nodes (e.g., NP subjects and direct objects).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Otherwise for the predecessor search hypothesis, we would have chosen a position that would not have been among the first n uncovered positions.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"95 76.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This section describes AdaBoost, which is the basis for the CoBoost algorithm.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Finally, quite a few hanzi are homographs, meaning that they may be pronounced in several different ways, and in extreme cases apparently represent different morphemes: The prenominal modifi\u00c2\u00ad cation marker eg deO is presumably a different morpheme from the second morpheme of \u00c2\u00a7eg mu4di4, even though they are written the same way.4 The second point, which will be relevant in the discussion of personal names in Section 4.4, relates to the internal structure of hanzi.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The question of what soft function to pick, and how to design' algorithms which optimize it, is an open question, but appears to be a promising way of looking at the problem.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"In focus is in particular the correlation with rhetorical structure, i.e., the question whether specific rhetorical relations \u00e2\u0080\u0094 or groups of relations in particular configurations \u00e2\u0080\u0094 are signalled by speakers with prosodic means.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Similar advances have been made in machine translation (Frederking and Nirenburg, 1994), speech recognition (Fiscus, 1997) and named entity recognition (Borthwick et al., 1998).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"To define a similarity function between the English and the foreign vertices, we rely on high-confidence word alignments.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The features are weighted within a logistic model to give an overall weight that is applied to the phrase pair\u2019s frequency prior to making MAP-smoothed relative-frequency estimates (different weights are learned for each conditioning direction).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Each round is composed of two stages; each stage updates one of the classifiers while keeping the other classifier fixed.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Process statistics are already collected by the kernel (and printing them has no meaningful impact on performance).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For eight judges, ranging k between 1 and 8 corresponded to a precision score range of 90% to 30%, meaning that there were relatively few words (30% of those found by the automatic segmenter) on which all judges agreed, whereas most of the words found by the segmenter were such that one human judge agreed.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"After adding a ROOT node to all trees, we train a grammar using six split-and- merge cycles and no Markovization.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The traveling salesman problem is an optimization problem which is defined as follows: given are a set of cities S = s1; ; sn and for each pair of cities si; sj the cost dij > 0 for traveling from city si to city sj . We are looking for the shortest tour visiting all cities exactly once while starting and ending in city s1.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We have modified Moses (Koehn et al., 2007) to keep our state with hypotheses; to conserve memory, phrases do not keep state.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"An initial step of any text\u00c2\u00ad analysis task is the tokenization of the input into words.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We will evaluate various specific aspects of the segmentation, as well as the overall segmentation per\u00c2\u00ad formance.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The scoping heuristics are based on the anaphor type: for reflexive pronouns the scope is the current clause, for relative pronouns it is the prior clause following its VP, for personal pronouns it is the anaphor\u00e2\u0080\u0099s sentence and two preceding sentences, and for definite NPs it is the anaphor\u00e2\u0080\u0099s sentence and eight preceding sentences.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The contextual role knowledge had the greatest impact on pronouns: +13% recall for terrorism and +15% recall for disasters, with a +1% precision gain in terrorism and a small precision drop of -3% in disasters.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"However, while Dienes and Dubey recognize empty categories in a pre-processing step and only let the parser find their antecedents, we use the parser both to detect dislocated dependents and to predict either the type or the location of their syntactic head (or both) and use post-processing only to transform the graph in accordance with the parser\u2019s analysis.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This FSA I can be segmented into words by composing Id(I) with D*, to form the WFST shown in Figure 2(c), then selecting the best path through this WFST to produce the WFST in Figure 2(d).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The method shares some characteristics of the decision list algorithm presented in this paper.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"encodes the one tag per word constraint and is uni form over type-level tag assignments.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"For comparison, the completely unsupervised feature-HMM baseline accuracy on the universal POS tags for English is 79.4%, and goes up to 88.7% with a treebank dictionary.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This task measures how well each package performs in machine translation.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The English data comes from the WSJ portion of the Penn Treebank and the other languages from the training set of the CoNLL-X multilingual dependency parsing shared task.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"To this end, we picked 100 sentences at random containing 4,372 total hanzi from a test corpus.14 (There were 487 marks of punctuation in the test sentences, including the sentence-final periods, meaning that the average inter-punctuation distance was about 9 hanzi.)","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"(S; C; j); Not only the coverage set C and the positions j; j0, but also the verbgroup states S; S0 are taken into account.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"It did 402 queries\/ms using 1.80 GB. cMemory use increased during scoring due to batch processing (MIT) or caching (Rand).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Parent Head Modif er Dir # gold F1 Label # gold F1 NP NP TAG R 946 0.54 ADJP 1216 59.45 S S S R 708 0.57 SBAR 2918 69.81 NP NP ADJ P R 803 0.64 FRAG 254 72.87 NP NP N P R 2907 0.66 VP 5507 78.83 NP NP SBA R R 1035 0.67 S 6579 78.91 NP NP P P R 2713 0.67 PP 7516 80.93 VP TAG P P R 3230 0.80 NP 34025 84.95 NP NP TAG L 805 0.85 ADVP 1093 90.64 VP TAG SBA R R 772 0.86 WHN P 787 96.00 S VP N P L 961 0.87 (a) Major phrasal categories (b) Major POS categories (c) Ten lowest scoring (Collins, 2003)-style dependencies occurring more than 700 times Table 8: Per category performance of the Berkeley parser on sentence lengths \u00e2\u0089\u00a4 70 (dev set, gold segmentation).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Search hypotheses are processed separately according to their coverage set C. The best scored hypothesis for each coverage set is computed: QBeam(C) = max e;e0 ;S;j Qe0 (e; S; C; j) The hypothesis (e0; e; S; C; j) is pruned if: Qe0 (e; S; C; j) < t0 QBeam(C); where t0 is a threshold to control the number of surviving hypotheses.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We realize the importance of paraphrase; however, the major obstacle is the construction of paraphrase knowledge.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Since pronouns carry little semantics of their own, resolving them depends almost entirely on context.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If somewhere else in the document we see \u00e2\u0080\u009crestrictions put in place by President Bush\u00e2\u0080\u009d, then we can be surer that Bush is a name.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For fair comparison, we have tabulated all results with the size of training data used (Table 5 and Table 6).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For that application, at a minimum, one would want to know the phonological word boundaries.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"As seen by the drop in average individual parser performance baseline, the introduced parser does not perform very well.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Email: rlls@bell-labs.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"suffixes (e.g., \u00ef\u00bf\u00bd=) Other notable parameters are second order vertical Markovization and marking of unary rules.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Table 3 Classes of words found by ST for the test corpus.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Our oracles took advantage of the labeled treebanks: While we tried to minimize the number of free parameters in our model, there are a few hyperparameters that need to be set.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This is because our corpus is not annotated, and hence does not distinguish between the various words represented by homographs, such as, which could be \/adv jiangl 'be about to' orInc jiang4 '(military) general'-as in 1j\\xiao3jiang4 'little general.'","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Performance improvements transfer to the Moses (Koehn et al., 2007), cdec (Dyer et al., 2010), and Joshua (Li et al., 2009) translation systems where our code has been integrated.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Experiments are presented in section 4.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Let us consider an example of ambiguity caused by devocalization.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Most languages that use Roman, Greek, Cyrillic, Armenian, or Semitic scripts, and many that use Indian-derived scripts, mark orthographic word boundaries; however, languages written in a Chinese-derived writ\u00c2\u00ad ing system, including Chinese and Japanese, as well as Indian-derived writing systems of languages like Thai, do not delimit orthographic words.1 Put another way, written Chinese simply lacks orthographic words.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"This allow the learners to &quot;bootstrap&quot; each other by filling the labels of the instances on which the other side has abstained so far.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Figure 2 shows timing results.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Zt can be written as follows Following the derivation of Schapire and Singer, providing that W+ > W_, Equ.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"A crucial difference is that the number of parameters is greatly reduced as is the number of variables that are sampled during each iteration.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"1 53.8 47.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Experiments are presented in section 4.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"It has no syntactic function.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This is appropriate in cases where it is sanctioned by Bayes\u2019 law, such as multiplying LM and TM probabilities, but for adaptation a more suitable framework is often a mixture model in which each event may be generated from some domain.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Firstly, Hebrew unknown tokens are doubly unknown: each unknown token may correspond to several segmentation possibilities, and each segment in such sequences may be able to admit multiple PoS tags.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"These make left-to-right query patterns convenient, as the application need only provide a state and the word to append, then use the returned state to append another word, etc.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The normalization factor plays an important role in the AdaBoost algorithm.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"A graph D = (W, A) is well-formed iff it is acyclic and connected.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Ties are rare in Bayes switching because the models are fine-grained \u2014 many estimated probabilities are involved in each decision.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The terms on the right-hand-side denote the type-level and token-level probability terms respectively.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The domains are the linguistic spans that are to receive an IS-partitioning, and the units are the (smaller) spans that can play a role as a constituent of such a partitioning.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"(1998) have also used a maximum entropy classifier that uses already tagged entities to help tag other entities.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"BerkeleyLM revision 152 (Pauls and Klein, 2011) implements tries based on hash tables and sorted arrays in Java with lossy quantization.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"DempsterShafer handles this by re-normalizing all the belief values with respect to only the non-null sets (this is the purpose of the denominator in Equation 1).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We present a coreference resolver called BABAR that uses contextual role knowledge to evaluate possible antecedents for an anaphor.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The hyperparameters \u00ce\u00b1 and \u00ce\u00b2 represent the concentration parameters of the token- and type-level components of the model respectively.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"REL+VB) (cf.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We use two common techniques, hash tables and sorted arrays, describing each before the model that uses the technique.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Our experimental setup therefore is designed to serve two goals.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"For the experiments in this paper we made a couple of additional modifications to the CoBoost algorithm.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Even with less training data, MENERGI outperforms Borthwick' s MENE + reference resolution (Borthwick, 1999).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"(2009) study related but different multilingual grammar and tagger induction tasks, where it is assumed that no labeled data at all is available.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"of Tokens MENERGI 318 160,000 200 180,000 IdentiFinder \u00e2\u0080\u0093 650,000 \u00e2\u0080\u0093 790,000 MENE \u00e2\u0080\u0093 \u00e2\u0080\u0093 350 321,000 Table 4: Training Data MUC7 test accuracy.2 For MUC6, the reduction in error due to global features is 27%, and for MUC7,14%.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"BABAR uses a named entity recognizer to identify proper names that refer to people and companies.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"In this particular case, all English vertices are labeled as nouns by the supervised tagger.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Evaluation Metrics We report three metrics to evaluate tagging performance.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Among these 32 sets, we found the following pairs of sets which have two or more links.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"1 1 0.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Annotation consistency is important in any supervised learning task.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"All annotations are done with specific tools and in XML; each layer has its own DTD.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"2.3 Assigning Evidence Values.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Lossy compressed models RandLM (Talbot and Osborne, 2007) and Sheffield (Guthrie and Hepple, 2010) offer better memory consumption at the expense of CPU and accuracy.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We weight edges using a unigram language model estimated with Good- Turing smoothing.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"3.1 Word ReOrdering with Verbgroup.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"There are two possible reasons: (1) the knowledge sources are resolving different cases of anaphora, and (2) the knowledge sources provide multiple pieces of evidence in support of (or against) a candidate, thereby acting synergistically to push the DempsterShafer model over the belief threshold in favor of a single candidate.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Finally, we note that Jiang\u2019s instance-weighting framework is broader than we have presented above, encompassing among other possibilities the use of unlabelled IN data, which is applicable to SMT settings where source-only IN corpora are available.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"83 77.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"pronunciations of individual words; they also need to compute intonational phrase boundaries in long utterances and assign relative prominence to words in those utterances.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Our original hope in combining these parsers is that their errors are independently distributed.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"There are two key benefits of this model architecture.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Clearly, explicitly modeling such a powerful constraint on tagging assignment has a potential to significantly improve the accuracy of an unsupervised part-of-speech tagger learned without a tagging dictionary.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In future work, we plan to follow-up on this approach and investigate other ways that contextual role knowledge can be used.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"So far the graph has been completely unlabeled.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(Specifically, the limit n starts at 5 and increases by 5 at each iteration.)","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":":zhong1 : 0.0 tjl :huo2 :0.0 (R:spub:\/ic of Ch:ina) + .,_,...I : jlong4 :0.0 (mUifaty genG181) 0 \u00c2\u00a3: _NC: 40.0 Figure 3 Partial Chinese Lexicon (NC = noun; NP = proper noun).c=- - I \u00e2\u0080\u00a2=- :il: .;ss:;zhangt \u00e2\u0080\u00a2 '-:.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"For each language under consideration, Petrov et al. (2011) provide a mapping A from the fine-grained language specific POS tags in the foreign treebank to the universal POS tags.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"On the one hand, the definition of composition in Steedman (1985), which technically permits composition of functions with unbounded number of arguments, generates tree sets with dependent paths such as those shown in Figure 6.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"diesem 3.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(7), such as the likelihood function used in maximum-entropy problems and other generalized additive models (Lafferty 99).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"These tag distributions are used to initialize the label distributions over the English vertices in the graph.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Since the parameter and token components will remain fixed throughout experiments, we briefly describe each.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Note that the backoff model assumes that there is a positive correlation between the frequency of a singular noun and its plural.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"By making use of global context, it has achieved excellent results on both MUC6 and MUC7 official test data.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"As an indication, in our core corpus, we found an average sentence length of 15.8 words and 1.8 verbs per sentence, whereas a randomly taken sample of ten commentaries from the national papers Su\u00c2\u00a8ddeutsche Zeitung and Frankfurter Allgemeine has 19.6 words and 2.1 verbs per sentence.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Segments with the same surface form but different PoS tags are treated as different lexemes, and are represented as separate arcs (e.g. the two arcs labeled neim from node 6 to 7).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This fully generative model caters for real interaction between the syntactic and morphological levels as a part of a single coherent process.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The type-level tag assignments T generate features associated with word types W . The tag assignments constrain the HMM emission parameters \u00ce\u00b8.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Simply assigning to each word its most frequent associated tag in a corpus achieves 94.6% accuracy on the WSJ portion of the Penn Treebank.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In previous work (Bean and Riloff, 1999), we developed an unsupervised learning algorithm that automatically recognizes definite NPs that are existential without syntactic modification because their meaning is universally understood.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"Here we present two algorithms.","target_text":"We present two algorithms.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"information structure.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Chang of Tsinghua University, Taiwan, R.O.C., for kindly providing us with the name corpora.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The dev corpus was taken from the NIST05 evaluation set, augmented with some randomly-selected material reserved from the training set.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The features we used can be divided into 2 classes: local and global.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We believe it is natural for authors to use abbreviations in subsequent mentions of a named entity (i.e., first \u00e2\u0080\u009cPresident George Bush\u00e2\u0080\u009d then \u00e2\u0080\u009cBush\u00e2\u0080\u009d).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"(2010) consistently outperforms ours on English, we obtain substantial gains across other languages.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The last affix in the list is the nominal plural f, men0.20 In the table are the (typical) classes of words to which the affix attaches, the number found in the test corpus by the method, the number correct (with a precision measure), and the number missed (with a recall measure).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We plan to explore more powerful techniques for exploiting the diversity of parsing methods.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Using the concept of inverted alignments, we explicitly take care of the coverage constraint by introducing a coverage set C of source sentence positions that have been already processed.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"2.1 Inverted Alignments.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Second, the reduced number of hidden variables and parameters dramatically speeds up learning and inference.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Also, in Information Extraction (IE), in which the system tries to extract elements of some events (e.g. date and company names of a corporate merger event), several event instances from different news articles have to be aligned even if these are expressed differently.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"To initialize the graph we tag the English side of the parallel text using a supervised model.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Others depend upon various lexical heuris\u00c2\u00ad tics: for example Chen and Liu (1992) attempt to balance the length of words in a three-word window, favoring segmentations that give approximately equal length for each word.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This would result in better rest cost estimation and better pruning.10 In general, tighter, but well factored, integration between the decoder and language model should produce a significant speed improvement.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"For robustness reasons, the parser may output a set of dependency trees instead of a single tree. most dependent of the next input token, dependency type features are limited to tokens on the stack.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Using this heuristic, BABAR identifies existential definite NPs in the training corpus using our previous learning algorithm (Bean and Riloff, 1999) and resolves all occurrences of the same existential NP with each another.1 2.1.2 Syntactic Seeding BABAR also uses syntactic heuristics to identify anaphors and antecedents that can be easily resolved.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"And if one is interested in TIS, one would probably consider the single orthographic word ACL to consist of three phonological words-lei s'i d\/-corresponding to the pronunciation of each of the letters in the acronym.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"What ought to be developed now is an annotation tool that can make use of the format, allow for underspecified annotations and visualize them accordingly.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This process produces a large set of caseframes coupled with a list of the noun phrases that they extracted.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"If they knew that the first four words in a hypergraph node would never extend to the left and form a 5-gram, then three or even fewer words could be kept in the backward state.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"So, it is too costly to make IE technology \u00e2\u0080\u009copen- domain\u00e2\u0080\u009d or \u00e2\u0080\u009con-demand\u00e2\u0080\u009d like IR or QA.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We are currently exploring such algorithms.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"As suggested in Section 4.3.2, a derivation with independent paths can be divided into subcomputations with limited sharing of information.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Then, the German infinitive 'besuchen' and the negation particle 'nicht' are translated.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"(f1; ;mg ; l) 2 (f1; ;mg n fl; l1g ; l0) !","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Non-empty buckets contain an entry belonging to them or to a preceding bucket where a conflict occurred.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The word joining is done on the basis of a likelihood criterion.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"It was then tested on section 22 of the Treebank in conjunction with the other parsers.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We use double-circles to indicate the space-delimited token boundaries.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In the case of, the most common usage is as an adverb with the pronunciation jiangl, so that variant is assigned the estimated cost of 5.98, and a high cost is assigned to nominal usage with the pronunciation jiang4.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"However, this argument is only plausible if the formal framework allows non-projective dependency structures, i.e. structures where a head and its dependents may correspond to a discontinuous constituent.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Named Entity Recognition: A Maximum Entropy Approach Using Global Information","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Several extensions of AdaBoost for multiclass problems have been suggested (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Hieu Hoang named the code \u201cKenLM\u201d and assisted with Moses along with Barry Haddow.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Similarly, hanzi sharing the GHOST radical _m tend to denote spirits and demons, such as _m gui3 'ghost' itself, II: mo2 'demon,' and yan3 'nightmare.'","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Therefore in cases where the segmentation is identical between the two systems we assume that tagging is also identical.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"D o m ai n Li n k ac cu ra cy W N c o v e r a g e C C 7 3 . 3 % 2 \/ 1 1 P C 8 8 . 9 % 2 \/ 8 Table 2.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"When OUT is large and distinct, its contribution can be controlled by training separate IN and OUT models, and weighting their combination.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This suggests a direct parallel to (1): where \u02dcp(s, t) is a joint empirical distribution extracted from the IN dev set using the standard procedure.2 An alternative form of linear combination is a maximum a posteriori (MAP) combination (Bacchiani et al., 2004).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Better Arabic Parsing: Baselines, Evaluations, and Analysis","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We call this approach parse hybridization.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Our approach uses word-to-word dependencies between source and target words.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"\u00e2\u0080\u00a2 Anaphoric links: the annotator is asked to specify whether the anaphor is a repetition, partial repetition, pronoun, epithet (e.g., Andy Warhol \u00e2\u0080\u0093 the PopArt artist), or is-a (e.g., Andy Warhol was often hunted by photographers.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"MADA uses an ensemble of SVMs to first re-rank the output of a deterministic morphological analyzer.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We used C = 1.0 as the L2 regularization constant in (Eq.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"3.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Word type N % Dic tion ary entr ies 2 , 5 4 3 9 7 . 4 7 Mor pho logi call y deri ved wor ds 3 0 . 1 1 Fore ign tran slite rati ons 9 0 . 3 4 Per son al na mes 5 4 2 . 0 7 cases.","label":0},{"source_text":"There is no global pruning.","target_text":"The 13 positions of the source sentence are processed in the order shown.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Specifically, we assume each word type W consists of feature-value pairs (f, v).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"88,962 (spelling,context) pairs were extracted as training data.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Assigning rhetorical relations thus poses questions that can often be answered only subjectively.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"2.2.4 Semantic Caseframe Expectations The third type of contextual role knowledge learned by BABAR is Semantic Caseframe Expectations.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Future work along these lines will incorporate other layers of annotation, in particular the syntax information.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Links can be of two different kinds: anaphoric or bridging (definite noun phrases picking up an antecedent via world-knowledge).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Initially, we planned to compare the semantic classes of an anaphor and a candidate and infer that they might be coreferent if their semantic classes intersected.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The translation direction is from German to English.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Our analysis begins with a description of syntactic ambiguity in unvocalized MSA text (\u00c2\u00a72).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"set was based on an earlier version of the Chang et a!.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Our test machine has two Intel Xeon E5410 processors totaling eight cores, 32 GB RAM, and four Seagate Barracuda disks in software RAID 0 running Linux 2.6.18.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Reference resolution involves finding words that co-refer to the same entity.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"It is also worth pointing out a connection with Daum\u00b4e\u2019s (2007) work that splits each feature into domain-specific and general copies.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Our experimental setup therefore is designed to serve two goals.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"So, who won the competition?","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Our results suggest that it is possible to learn accurate POS taggers for languages which do not have any annotated data, but have translations into a resource-rich language.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Other errors include NE tagging errors and errors due to a phrase which includes other NEs.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"For each experiment we gave an nonparametric and a parametric technique for combining parsers.","label":0},{"source_text":"There is no global pruning.","target_text":"An error count of 0:0 is assigned to a perfect translation, and an error count of 1:0 is assigned to a semantically and syntactically wrong translation.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"It is unique, agrees with the maximum-likelihood distribution, and has the exponential form (Della Pietra et al., 1997): where refers to the outcome, the history (or context), and is a normalization function.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"32 81.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"As with personal names, we also derive an estimate from text of the probability of finding a transliterated name of any kind (PTN).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"(2009).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Nu mb er filters candidate if number doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"It is likely that the more complex cases, where path information could make a difference, are beyond the reach of the parser in most cases.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Due to the dramatic fiscal situation in Brandenburg she now surprisingly withdrew legislation drafted more than a year ago, and suggested to decide on it not before 2003.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Model components cascade, so the row corresponding to +FEATS also includes the PRIOR component (see Section 3).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"While sorted arrays could be used to implement the same data structure as PROBING, effectively making m = 1, we abandoned this implementation because it is slower and larger than a trie implementation.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This design does not guarantee \u00e2\u0080\u009cstructural zeros,\u00e2\u0080\u009d but biases towards sparsity.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The availability of these resources guided our selection of foreign languages.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We used it to score all phrase pairs in the OUT table, in order to provide a feature for the instance-weighting model.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We formulate the update as follows: where \u2200ui \u2208 Vf \\ Vfl, \u03b3i(y) and \u03bai are defined as: We ran this procedure for 10 iterations.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In the pinyin transliterations a dash(-) separates syllables that may be considered part of the same phonological word; spaces are used to separate plausible phonological words; and a plus sign (+) is used, where relevant, to indicate morpheme boundaries of interest.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"All three curves remain steep at the maximum training set size of 18818 trees.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Morphological Analyzer Ideally, we would use an of-the-shelf morphological analyzer for mapping each input token to its possible analyses.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"This process produces a large set of caseframes coupled with a list of the noun phrases that they extracted.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Combining multiple highly-accurate independent parsers yields promising results.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Jud ges A G G R ST M 1 M 2 M 3 T1 T2 T3 AG 0.7 0 0.7 0 0 . 4 3 0.4 2 0.6 0 0.6 0 0.6 2 0.5 9 GR 0.9 9 0 . 6 2 0.6 4 0.7 9 0.8 2 0.8 1 0.7 2 ST 0 . 6 4 0.6 7 0.8 0 0.8 4 0.8 2 0.7 4 M1 0.7 7 0.6 9 0.7 1 0.6 9 0.7 0 M2 0.7 2 0.7 3 0.7 1 0.7 0 M3 0.8 9 0.8 7 0.8 0 T1 0.8 8 0.8 2 T2 0.7 8 respectively, the recall and precision.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The same argument holds for resolving PP attachment of a prefixed preposition or marking conjunction of elements of any kind.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Thus it is possible, for illustration, to look for a noun phrase (syntax tier) marked as topic (information structure tier) that is in a bridging relation (co-reference tier) to some other noun phrase.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In this section we present a partial evaluation of the current system, in three parts.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The number of top-ranked pairs to retain is chosen to optimize dev-set BLEU score.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"2.1 Part-of-speech tags.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The Wang, Li, and Chang system fails on fragment (b) because their system lacks the word youlyoul 'soberly' and misinterpreted the thus isolated first youl as being the final hanzi of the preceding name; similarly our system failed in fragment (h) since it is missing the abbreviation i:lJI!","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"ATB CTB6 Negra WSJ Trees 23449 28278 20602 43948 Word Typess 40972 45245 51272 46348 Tokens 738654 782541 355096 1046829 Tags 32 34 499 45 Phrasal Cats 22 26 325 27 Test OOV 16.8% 22.2% 30.5% 13.2% Per Sentence Table 4: Gross statistics for several different treebanks.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Two aspects of the corpus have been presented in previous papers ((Re- itter, Stede 2003) on underspecified rhetorical structure; (Stede 2003) on the perspective of knowledge-based summarization).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This Good\u00c2\u00ad Turing estimate of p(unseen(f,) If,) can then be used in the normal way to define the probability of finding a novel instance of a construction in ir, in a text: p(unseen(f,)) = p(unseen(f,) I f,) p(fn Here p(ir,) is just the probability of any construction in ft as estimated from the frequency of such constructions in the corpus.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Morphological analyzers for Hebrew that analyze a surface form in isolation have been proposed by Segal (2000), Yona and Wintner (2005), and recently by the knowledge center for processing Hebrew (Itai et al., 2006).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"MENE (Maximum Entropy Named Entity) (Borth- wick, 1999) was combined with Proteus (a hand- coded system), and came in fourth among all MUC 7 participants.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"While processing the source sentence monotonically, the initial state I is entered whenever there are no uncovered positions to the left of the rightmost covered position.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"As already pointed out in Section 2.4, current theories diverge not only on the number and definition of relations but also on apects of structure, i.e., whether a tree is sufficient as a representational device or general graphs are required (and if so, whether any restrictions can be placed on these graph\u00e2\u0080\u0099s structures \u00e2\u0080\u0094 cf.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"A similar structure is used in speech recognition.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(1998) did make use of information from the whole document.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"(2010) consistently outperforms ours on English, we obtain substantial gains across other languages.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Particles are uninflected.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"If a phrase does not contain any keywords, the phrase is discarded.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"First, from a large corpus, we extract all the NE instance pairs.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, even if the contexts surrounding an anaphor and candidate match exactly, they are not coreferent if they have substantially different meanings 9 We would be happy to make our manually annotated test data available to others who also want to evaluate their coreference resolver on the MUC4 or Reuters collections.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"rently, some annotations (in particular the connectives and scopes) have already moved beyond the core corpus; the others will grow step by step.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Two aspects of the corpus have been presented in previous papers ((Re- itter, Stede 2003) on underspecified rhetorical structure; (Stede 2003) on the perspective of knowledge-based summarization).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"It also uses less memory, with 8 bytes of overhead per entry (we store 16-byte entries with m = 1.5); linked list implementations hash set and unordered require at least 8 bytes per entry for pointers.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The CoBoost algorithm just described is for the case where there are two labels: for the named entity task there are three labels, and in general it will be useful to generalize the CoBoost algorithm to the multiclass case.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The scoping heuristics are based on the anaphor type: for reflexive pronouns the scope is the current clause, for relative pronouns it is the prior clause following its VP, for personal pronouns it is the anaphor\u00e2\u0080\u0099s sentence and two preceding sentences, and for definite NPs it is the anaphor\u00e2\u0080\u0099s sentence and eight preceding sentences.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This class-based model gives reasonable results: for six radical classes, Table 1 gives the estimated cost for an unseen hanzi in the class occurring as the second hanzi in a double GIVEN name.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Clearly, explicitly modeling such a powerful constraint on tagging assignment has a potential to significantly improve the accuracy of an unsupervised part-of-speech tagger learned without a tagging dictionary.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Preferences for constituent order (especially in languages with relatively free word order) often belong to this group.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Next, for each pair of NE categories, we collect all the contexts and find the keywords which are topical for that NE category pair.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Thus we opted not to take the step of creating more precise written annotation guidelines (as (Carlson, Marcu 2001) did for English), which would then allow for measuring inter-annotator agreement.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"This feature has a linguistic justification.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"This approach is compared to another reordering scheme presented in (Berger et al., 1996).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Unlike Germann et al. (2009), we chose a model size so that all benchmarks fit comfortably in main memory.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"An anti-greedy algorithm, AG: instead of the longest match, take the.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Figure 2 shows timing results.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"In contrast, NNP (proper nouns) form a large portion of vocabulary.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We obtained positive results using a very simple phrase-based system in two different adaptation settings: using English\/French Europarl to improve a performance on a small, specialized medical domain; and using non-news portions of the NIST09 training material to improve performance on the news-related corpora.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Call the crossing constituents A and B.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"At most one feature in this group will be set to 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"A summary of the corpus used in the experiments is given in Table 3.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"MENE (without the help of hand-coded systems) has been shown to be somewhat inferior in performance.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We use graph-based label propagation for cross-lingual knowledge transfer and use the projected labels as features in an unsupervised model (Berg- Kirkpatrick et al., 2010).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Hash tables are a common sparse mapping technique used by SRILM\u2019s default and BerkeleyLM\u2019s hashed variant.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"In the label propagation stage, we propagate the automatic English tags to the aligned Italian trigram types, followed by further propagation solely among the Italian vertices. the Italian vertices are connected to an automatically labeled English vertex.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"A more forceful approach for encoding sparsity is posterior regularization, which constrains the posterior to have a small number of expected tag assignments (Grac\u00c2\u00b8a et al., 2009).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"For comparison, the completely unsupervised feature-HMM baseline accuracy on the universal POS tags for English is 79.4%, and goes up to 88.7% with a treebank dictionary.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"(7) \u03c6 s,t This is a somewhat less direct objective than used by Matsoukas et al, who make an iterative approximation to expected TER.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Unfortunately, modifying the model to account for these kind of dependencies is not at all straightforward.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Evaluation results within sets Table 1 shows the evaluation result based on the number of phrases in a set.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Hence, for each token, one of the four features zone-TXT, zone- HL, zone-DATELINE, or zone-DD is set to 1, and the other 3 are set to 0.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The bias of automatic methods in favor of statistical systems seems to be less pronounced on out-of-domain test data.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The first four affixes are so-called resultative affixes: they denote some prop\u00c2\u00ad erty of the resultant state of a verb, as in E7 wang4bu4-liao3 (forget-not-attain) 'cannot forget.'","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR applies a DempsterShafer probabilistic model to make resolutions based on evidence from the contextual role knowledge sources as well as general knowledge sources.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The interdependence between fb or 1\/!i, and 5:2 is not captured by our model, but this could easily be remedied.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"There is no global pruning.","target_text":"Our new approach, which is called quasi-monotone search, processes the source sentence monotonically, while explicitly taking into account the positions of the German verbgroup.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"4.4 Chinese Personal Names.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR employs information extraction techniques to represent and learn role relationships.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We thank members of the MIT NLP group for their suggestions and comments.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Thus at each iteration the algorithm is forced to pick features for the location, person and organization in turn for the classifier being trained.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Clearly this poses a number of research challenges, though, such as the applicability of tag sets across different languages.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Evaluation results for links","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In MSA, SVO usually appears in non-matrix clauses.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(4) is minimized by setting Since a feature may be present in only a few examples, W_ can be in practice very small or even 0, leading to extreme confidence values.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"One striking example is the error reduction for Spanish, which reduces error by 36.5% and 24.7% for the best and median settings respectively.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Je voudrais pr\u00b4eciser, a` l\u2019adresse du commissaire Liikanen, qu\u2019il n\u2019est pas ais\u00b4e de recourir aux tribunaux nationaux.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"What both of these approaches presume is that there is a sin\u00c2\u00ad gle correct segmentation for a sentence, against which an automatic algorithm can be compared.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"To conserve memory at the expense of accuracy, values may be quantized using q bits per probability and r bits per backoff6.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"19 We note that it is not always clear in Wang, Li, and Chang's examples which segmented words.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"1 1 0.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The ratio of buckets to entries is controlled by space multiplier m > 1.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"La ng ua ge 1T W + P RI O R + F E A T S E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 2 1.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"This is manifest in the lexical choices but 1 www.coli.unisb.de\/\u00e2\u0088\u00bcthorsten\/tnt\/ Dagmar Ziegler is up to her neck in debt.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"\u00e2\u0080\u009d).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(a) I f f fi * fi :1 }'l ij 1\u00c2\u00a7: {1M m m s h e n 3 m e 0 shi2 ho u4 wo 3 cai2 ne ng 2 ke4 fu 2 zh e4 ge 4 ku n4 w h a t ti m e I just be abl e ov er co m e thi s C L dif fic 'When will I be able to overcome this difficulty?'","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"One obvious application is information extraction.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Each trie node contains a sorted array of entries and they use binary search.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This is similar to using the Linux MAP POPULATE flag that is our default loading mechanism.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Global features are extracted from other occurrences of the same token in the whole document.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"3.1 Gross Statistics.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In general, m; l; l0 6= fl1; l2; l3g and in line umber 3 and 4, l0 must be chosen not to violate the above reordering restriction.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"(2009).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This is manifest in the lexical choices but 1 www.coli.unisb.de\/\u00e2\u0088\u00bcthorsten\/tnt\/ Dagmar Ziegler is up to her neck in debt.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"99 94.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"2.5 Connectives with scopes.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Hereafter, each pair of NE categories will be called a domain; e.g. the \u00e2\u0080\u009cCompany \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain, which we will call CC- domain (Step 2).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Unlike the WSJ corpus which has a high frequency of rules like VP \u00e2\u0086\u0092VB PP, Arabic verb phrases usually have lexi calized intervening nodes (e.g., NP subjects and direct objects).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"For English POS tagging, BergKirkpatrick et al. (2010) found that this direct gradient method performed better (>7% absolute accuracy) than using a feature-enhanced modification of the Expectation-Maximization (EM) algorithm (Dempster et al., 1977).8 Moreover, this route of optimization outperformed a vanilla HMM trained with EM by 12%.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"3.2 Inter-annotator Agreement.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"For the \u00e2\u0080\u0098core\u00e2\u0080\u0099 portion of PCC, we found that on average, 35% of the coherence relations in our RST annotations are explicitly signalled by a lexical connective.6 When adding the fact that connectives are often ambiguous, one has to conclude that prospects for an automatic analysis of rhetorical structure using shallow methods (i.e., relying largely on connectives) are not bright \u00e2\u0080\u0094 but see Sections 3.2 and 3.3 below.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We tagged each noun with the top-level semantic classes assigned to it in Word- Net.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"First, we use a novel graph-based framework for projecting syntactic information across language boundaries.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"One key issue here is to seek a discourse-based model of information structure.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Only IRSTLM does not support threading.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As a first step towards modeling transliterated names, we have collected all hanzi occurring more than once in the roughly 750 foreign names in our dictionary, and we estimate the probabil\u00c2\u00ad ity of occurrence of each hanzi in a transliteration (pTN(hanzi;)) using the maximum likelihood estimate.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"47 78.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Using the terminology of Kahane et al. (1998), we say that jedna is the syntactic head of Z, while je is its linear head in the projectivized representation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For example, hanzi containing the INSECT radical !R tend to denote insects and other crawling animals; examples include tr wal 'frog,' feng1 'wasp,' and !Itt she2 'snake.'","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Graph construction does not require any labeled data, but makes use of two similarity functions.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For example, we can easily imagine that the number of paraphrases for \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d is enormous and it is not possible to create a comprehensive inventory by hand.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"This design does not guarantee \u00e2\u0080\u009cstructural zeros,\u00e2\u0080\u009d but biases towards sparsity.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"By considering derivation trees, and thus abstracting away from the details of the composition operation and the structures being manipulated, we are able to state the similarities and differences between the 'This work was partially supported by NSF grants MCS42-19116-CER, MCS82-07294 and DCR-84-10413, ARO grant DAA 29-84-9-0027, and DARPA grant N00014-85-K0018.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In many cases, there is an even stronger restriction: over large portions of the source string, the alignment is monotone.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Fortunately, there are only a few hundred hanzi that are particularly common in transliterations; indeed, the commonest ones, such as E. bal, m er3, and iij al are often clear indicators that a sequence of hanzi containing them is foreign: even a name like !:i*m xia4mi3-er3 'Shamir,' which is a legal Chi\u00c2\u00ad nese personal name, retains a foreign flavor because of liM.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Named Entity Recognition: A Maximum Entropy Approach Using Global Information","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(4) is minimized by setting Since a feature may be present in only a few examples, W_ can be in practice very small or even 0, leading to extreme confidence values.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Several papers report the use of part-of-speech information to rank segmentations (Lin, Chiang, and Su 1993; Peng and Chang 1993; Chang and Chen 1993); typically, the probability of a segmentation is multiplied by the probability of the tagging(s) for that segmentation to yield an estimate of the total probability for the analysis.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"There has been a lot of research on such lexical relations, along with the creation of resources such as WordNet.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Step 1.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We evaluate our approach on seven languages: English, Danish, Dutch, German, Portuguese, Spanish, and Swedish.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"As can be seen from the last column in Table 1, both Head and Head+Path may theoretically lead to a quadratic increase in the number of distinct arc labels (Head+Path being worse than Head only by a constant factor), while the increase is only linear in the case of Path.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In addition, each feature function is a binary function.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Specifically, we (+FEATS) utilizes the tag prior as well as features (e.g., suffixes and orthographic features), discussed in Section 3, for the P (W |T , \u00cf\u0088) component.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Using this heuristic, BABAR identifies existential definite NPs in the training corpus using our previous learning algorithm (Bean and Riloff, 1999) and resolves all occurrences of the same existential NP with each another.1 2.1.2 Syntactic Seeding BABAR also uses syntactic heuristics to identify anaphors and antecedents that can be easily resolved.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Also, we don\u00e2\u0080\u0099t know how many such paraphrase sets are necessary to cover even some everyday things or events.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The problem of &quot;noise&quot; items that do not fall into any of the three categories also needs to be addressed.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Phrase-level granularity distinguishes our work from previous work by Matsoukas et al (2009), who weight sentences according to sub-corpus and genre membership.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We propose a limit of 70 words for Arabic parsing evaluations.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"2 56.2 32.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The verbal reading arises when the maSdar has an NP argument which, in vocalized text, is marked in the accusative case.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"This group consists of (3 total number of possible zones) features.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The choice of the genre commentary resulted from the fact that an investigation of rhetorical structure, its interaction with other aspects of discourse structure, and the prospects for its automatic derivation are the key motivations for building up the corpus.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"This group contains a large number of features (one for each token string present in the training data).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The maximum entropy framework estimates probabilities based on the principle of making as few assumptions as possible, other than the constraints imposed.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This approach needs a phrase as an initial seed and thus the possible relationships to be extracted are naturally limited.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"We apply a beam search concept as in speech recognition.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"3.2 The DempsterShafer Decision Model.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Matsoukas et al (2009) generalize it by learning weights on sentence pairs that are used when estimating relative-frequency phrase-pair probabilities.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"However, lazy mapping is generally slow because queries against uncached pages must wait for the disk.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"In fact, we found that enabling IRSTLM\u2019s cache made it slightly slower, so results in Table 1 use IRSTLM without caching.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"M spawns as many processes as there are ways of breaking up ri , .. \u2022 , zt, and rules with A on their left-hand-side.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"As with HG's derivation structures are annotated; in the case of TAG's, by the trees used for adjunction and addresses of nodes of the elementary tree where adjunctions occurred.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"49 99.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Ex: The regime gives itself the right...","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The simplest approach involves scoring the various analyses by costs based on word frequency, and picking the lowest cost path; variants of this approach have been described in Chang, Chen, and Chen (1991) and Chang and Chen (1993).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We focus on phrases which connect two Named Entities (NEs), and proceed in two stages.","label":0},{"source_text":"There is no global pruning.","target_text":"Again, the monotone search performs worst.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"This remains as future work.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In the experiments below, we employ a data-driven deterministic dependency parser producing labeled projective dependency graphs,3 previously tested on Swedish (Nivre et al., 2004) and English (Nivre and Scholz, 2004).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"\u00ce\u00b8 has a belief value of 1.0, indicating complete certainty that the correct hypothesis is included in the set, and a plausibility value of 1.0, indicating that there is no evidence for competing hypotheses.5 As evidence is collected and the likely hypotheses are whittled down, belief is redistributed to subsets of \u00ce\u00b8.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Thus the method makes the fairly strong assumption that the features can be partitioned into two types such that each type alone is sufficient for classification.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Before we turn to the evaluation, however, we need to introduce the data-driven dependency parser used in the latter experiments.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Table 2 compares the performance of our system on the setup of Cohen and Smith (2007) to the best results reported by them for the same tasks.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Since MUC6, BBN' s Hidden Markov Model (HMM) based IdentiFinder (Bikel et al., 1997) has achieved remarkably good performance.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"In our particular case the majority requires the agreement of only two parsers because we have only three.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(8) can now be rewritten5 as which is of the same form as the function Zt used in AdaBoost.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"This work was supported in part by the National Science Foundation under grant IRI9704240.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For previously unseen hanzi in given names, Chang et al. assign a uniform small cost; but we know that some unseen hanzi are merely acci\u00c2\u00ad dentally missing, whereas others are missing for a reason-for example, because they have a bad connotation.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Unfortunately, we were unable to correctly run the IRSTLM quantized variant.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"As can be seen in Figure 3, the phrases in the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set include completely different relationships, which are not paraphrases.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"(f1; ;mg n fl1g ; l) 3 (f1; ;mg n fl; l1; l2g ; l0) !","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The P (T |\u00cf\u0088) distribution, in English for instance, should have very low mass for the DT (determiner) tag, since determiners are a very small portion of the vocabulary.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Thus, the effects of spontaneous speech are present in the corpus, e.g. the syntactic structure of the sentence is rather less restricted, however the effect of speech recognition errors is not covered.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We then discuss how we adapt and generalize a boosting algorithm, AdaBoost, to the problem of named entity classification.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Another attempt at using global information can be found in (Borthwick, 1999).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The number of latent HMM states for each language in our experiments was set to the number of fine tags in the language\u2019s treebank.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Compared to decoding, this task is cache-unfriendly in that repeated queries happen only as they naturally occur in text.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Of course, we.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"IRSTLM and BerkeleyLM use this state function (and a limit of N \u22121 words), but it is more strict than necessary, so decoders using these packages will miss some recombination opportunities.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The effect of a second reference resolution classifier is not entirely the same as that of global features.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This suggests a direct parallel to (1): where \u02dcp(s, t) is a joint empirical distribution extracted from the IN dev set using the standard procedure.2 An alternative form of linear combination is a maximum a posteriori (MAP) combination (Bacchiani et al., 2004).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Many researchers have developed coreference resolvers, so we will only discuss the methods that are most closely related to BABAR.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"However, their inverted variant implements a reverse trie using less CPU and the same amount of memory7.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"7).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"As described in Sproat (1995), the Chinese segmenter presented here fits directly into the context of a broader finite-state model of text analysis for speech synthesis.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"It reads a file with a list of German connectives, and when a text is opened for annotation, it highlights all the words that show up in this list; these will be all the potential connectives.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Clearly it is possible to write a rule that states that if an analysis Modal+ Verb is available, then that is to be preferred over Noun+ Verb: such a rule could be stated in terms of (finite-state) local grammars in the sense of Mohri (1993).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Because of this threshold, very few NE instance pairs could be used and hence the variety of phrases was also limited.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"\u00e2\u0080\u00a2 Some tools would allow for the desired annotation mode, but are so complicated (they can be used for many other purposes as well) that annotators take a long time getting used to them.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"They are: 5We are grateful to an anonymous reviewer for pointing this out.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"However, it is unlikely that other occurrences of News Broadcasting Corp. in the same document also co-occur with Even.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Ablation Analysis We evaluate the impact of incorporating various linguistic features into our model in Table 3.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Graph construction for structured prediction problems such as POS tagging is non-trivial: on the one hand, using individual words as the vertices throws away the context necessary for disambiguation; on the other hand, it is unclear how to define (sequence) similarity if the vertices correspond to entire sentences.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"In the interest of testing the robustness of these combining techniques, we added a fourth, simple nonlexicalized PCFG parser.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Table 5 provides insight into the behavior of different models in terms of the tagging lexicon they generate.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For previously unseen hanzi in given names, Chang et al. assign a uniform small cost; but we know that some unseen hanzi are merely acci\u00c2\u00ad dentally missing, whereas others are missing for a reason-for example, because they have a bad connotation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In many cases these failures in recall would be fixed by having better estimates of the actual prob\u00c2\u00ad abilities of single-hanzi words, since our estimates are often inflated.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The development set contained 44088 constituents in 2416 sentences and the test set contained 30691 constituents in 1699 sentences.","label":0},{"source_text":"There is no global pruning.","target_text":"Depending on the threshold t0, the search algorithm may miss the globally optimal path which typically results in additional translation errors.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"Here we present two algorithms.","target_text":"An edge indicates that the two features must have the same label.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(In this figure eps is c) be implemented, though, such as a maximal-grouping strategy (as suggested by one reviewer of this paper); or a pairwise-grouping strategy, whereby long sequences of unattached hanzi are grouped into two-hanzi words (which may have some prosodic motivation).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Second, BABAR performs reliable case resolution to identify anaphora that can be easily resolved using the lexical and syntactic heuristics described in Section 2.1.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Person\u00e2\u0080\u009d domain, 618 keywords are found, and in the \u00e2\u0080\u009cCountry \u00e2\u0080\u0093 Country\u00e2\u0080\u009d domain, 303 keywords are found.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The contextual role knowledge had the greatest impact on pronouns: +13% recall for terrorism and +15% recall for disasters, with a +1% precision gain in terrorism and a small precision drop of -3% in disasters.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"7 Big 5 is the most popular Chinese character coding standard in use in Taiwan and Hong Kong.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"52 15.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"There, a lattice is used to represent the possible sentences resulting from an interpretation of an acoustic model.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"F1 85 Berkeley 80 Stanford.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"9 50.2 +P RI OR be st me dia n 47.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The 1st block contains the simple baselines from section 2.1.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Finally, the concatenated 5 * 20% output is used to train the reference resolution component.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Taking \/V, to be the number of examples an algorithm classified correctly (where all gold standard items labeled noise were counted as being incorrect), we calculated two measures of accuracy: See Tab.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Specifically, the lexicon is generated as: P (T , W |\u00cf\u0088) =P (T )P (W |T ) Word Type Features (FEATS): Past unsupervised POS work have derived benefits from features on word types, such as suffix and capitalization features (Hasan and Ng, 2009; Berg-Kirkpatrick et al.,2010).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We can represent any derived tree of a TAG by the two substrings that appear in its frontier, and then define how the adjunction operation concatenates the substrings.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Clearly, for judges h and h taking h as standard and computing the precision and recall for Jz yields the same results as taking h as the standard, and computing for h, 14 All evaluation materials, with the exception of those used for evaluating personal names were drawn.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This has the potential drawback of increasing the number of features, which can make MERT less stable (Foster and Kuhn, 2009).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"An inverted alignment is defined as follows: inverted alignment: i ! j = bi: Target positions i are mapped to source positions bi.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The natural baseline approach is to concatenate data from IN and OUT.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Finally, we intend to explore more sophisticated instanceweighting features for capturing the degree of generality of phrase pairs.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In general, different modalities (\u00e2\u0080\u009cplanned to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cagreed to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cbought\u00e2\u0080\u009d) were considered to express the same relationship within an extraction setting.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"MENE is then trained on 80% of the training corpus, and tested on the remaining 20%.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The distribution specifies the relative weight, or importance, of each example \u2014 typically, the weak learner will attempt to minimize the weighted error on the training set, where the distribution specifies the weights.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"This paper does not necessarily reflect the position of the U.S. Government.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We use MMAX for this annotation as well.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The segmenter will give both analyses :1 cai2 neng2 'just be able,' and ?]cai2neng2 'talent,' but the latter analysis is preferred since splitting these two morphemes is generally more costly than grouping them.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The focus of our work is on the use of contextual role knowledge for coreference resolution.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We focus here instead on adapting the two most important features: the language model (LM), which estimates the probability p(wIh) of a target word w following an ngram h; and the translation models (TM) p(slt) and p(t1s), which give the probability of source phrase s translating to target phrase t, and vice versa.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Cohen and Smith (2007) chose a metric like SParseval (Roark et al., 2006) that first aligns the trees and then penalizes segmentation errors with an edit-distance metric.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Besides information structure, the second main goal is to enhance current models of rhetorical structure.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Arguably this consists of about three phonological words.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"10) and trained both EM and L-BFGS for 1000 iterations.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Table 2 shows our complete set of results.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As described in Sproat (1995), the Chinese segmenter presented here fits directly into the context of a broader finite-state model of text analysis for speech synthesis.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"2 60.7 56.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"For a sentence x and a state sequence z, a first order Markov model defines a distribution: (9) where Val(X) corresponds to the entire vocabulary.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR merely identifies caseframes that frequently co-occur in coreference resolutions.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Secondly, for all our models we provide better fine- and coarse-grained POS-tagging accuracy, and all pruned models outperform the Oracle results reported by them.12 In terms of syntactic disambiguation, even the simplest grammar pruned with HSPELL outperforms their non-Oracle results.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Table 2 shows results for both settings and all methods described in sections 2 and 3.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Table 1 shows four words \u00e2\u0080\u009c 0 Indeed NN Indeed Saddamwhose unvocalized surface forms 0 an are indistinguishable.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We feel that information from a whole corpus might turn out to be noisy if the documents in the corpus are not of the same genre.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We run the baseline Moses system for the French-English track of the 2011 Workshop on Machine Translation,9 translating the 3003-sentence test set.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Generalizing state minimization, the model could also provide explicit bounds on probability for both backward and forward extension.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"An error count of 0:0 is assigned to a perfect translation, and an error count of 1:0 is assigned to a semantically and syntactically wrong translation.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.3 Rhetorical structure.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In Table 5 we present results from small test cor\u00c2\u00ad pora for the productive affixes handled by the current version of the system; as with names, the segmentation of morphologically derived words is generally either right or wrong.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Each of the constituents must have received at least 1 votes from the k parsers, so a > I1 and 2 \u2014 2k\u00b1-1 b > ri-5-111.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"It is difficult to directly compare the Matsoukas et al results with ours, since our out-of-domain corpus is homogeneous; given heterogeneous training data, however, it would be trivial to include Matsoukas-style identity features in our instance-weighting model.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"conceptual relationship in the discourse.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As we have noted in Section 2, the general semantic class to which a hanzi belongs is often predictable from its semantic radical.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Using the virtual distribution Di (i) and pseudo-labels&quot;y.,\u201e values for Wo, W\u00b1 and W_ can be calculated for each possible weak hypothesis (i.e., for each feature x E Xi); the weak hypothesis with minimal value for Wo + 2\/WW _ can be chosen as before; and the weight for this weak hypothesis at = ln ww+411:) can be calculated.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"One key issue here is to seek a discourse-based model of information structure.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"For the translation experiments, Eq. 2 is recursively evaluated.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"4.1 Dictionary Representation.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Email: rlls@bell-labs.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"(2010) and the posterior regular- ization HMM of Grac\u00c2\u00b8a et al.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The above auxiliary quantity satisfies the following recursive DP equation: Qe0 (e; C; j) = Initial Skip Verb Final 1.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Our baseline for all sentence lengths is 5.23% F1 higher than the best previous result.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Unique Occurrences and Zone (UNIQ): This group of features indicates whether the word is unique in the whole document.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"0 57.2 43.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The choice of the particular newspaper was motivated by the fact that the language used in a regional daily is somewhat simpler than that of papers read nationwide.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"During testing, it is possible that the classifier produces a sequence of inadmissible classes (e.g., person begin followed by location unique).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Each feature concept is akin to a random variable and its occurrence in the text corresponds to a particular instantiation of that random variable.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"But it conflates the coordinating and discourse separator functions of wa (<..4.b \u00ef\u00bf\u00bd \u00ef\u00bf\u00bd) into one analysis: conjunction(Table 3).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For terrorism, BABAR generated 5,078 resolutions: 2,386 from lexical seeding and 2,692 from syntactic seeding.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For example, in the CC-domain, 96 keywords are found which have TF\/ITF scores above a threshold; some of them are shown in Figure 3.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The weight on each sentence is a value in [0, 1] computed by a perceptron with Boolean features that indicate collection and genre membership.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The idea of distinguishing between general and domain-specific examples is due to Daum\u00b4e and Marcu (2006), who used a maximum-entropy model with latent variables to capture the degree of specificity.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Full Chinese personal names are in one respect simple: they are always of the form family+given.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"In this paper we study the problem of using a parallel corpus from a background domain (OUT) to improve performance on a target domain (IN) for which a smaller amount of parallel training material\u2014though adequate for reasonable performance\u2014is also available.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Equ.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"For the \u00e2\u0080\u0098core\u00e2\u0080\u0099 portion of PCC, we found that on average, 35% of the coherence relations in our RST annotations are explicitly signalled by a lexical connective.6 When adding the fact that connectives are often ambiguous, one has to conclude that prospects for an automatic analysis of rhetorical structure using shallow methods (i.e., relying largely on connectives) are not bright \u00e2\u0080\u0094 but see Sections 3.2 and 3.3 below.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"During testing, it is possible that the classifier produces a sequence of inadmissible classes (e.g., person begin followed by location unique).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Somewhat surprisingly, there do not appear to be large systematic differences between linear and MAP combinations.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Since foreign names can be of any length, and since their original pronunciation is effectively unlimited, the identi\u00c2\u00ad fication of such names is tricky.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"), which precludes a single universal approach to adaptation.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For a language like English, this problem is generally regarded as trivial since words are delimited in English text by whitespace or marks of punctuation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This latter evaluation compares the performance of the system with that of several human judges since, as we shall show, even people do not agree on a single correct way to segment a text.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Table 2 shows single-threaded results, mostly for comparison to IRSTLM, and Table 3 shows multi-threaded results.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In contrast to the Bayesian HMM, \u00ce\u00b8t is not drawn from a distribution which has support for each of the n word types.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In the maximum entropy framework, there is no such constraint.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This extends previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and using a simpler training procedure.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Note that while the standard HMM, has O(K n) emission parameters, our model has O(n) effective parameters.3 Token Component Once HMM parameters (\u00cf\u0086, \u00ce\u00b8) have been drawn, the HMM generates a token-level corpus w in the standard way: P (w, t|\u00cf\u0086, \u00ce\u00b8) = P (T , W , \u00ce\u00b8, \u00cf\u0088, \u00cf\u0086, t, w|\u00ce\u00b1, \u00ce\u00b2) = P (T , W , \u00cf\u0088|\u00ce\u00b2) [Lexicon] \u00ef\u00a3\u00ab n n \u00ef\u00a3\u00ad (w,t)\u00e2\u0088\u0088(w,t) j \u00ef\u00a3\u00b6 P (tj |\u00cf\u0086tj\u00e2\u0088\u00921 )P (wj |tj , \u00ce\u00b8tj )\u00ef\u00a3\u00b8 P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) [Parameter] P (w, t|\u00cf\u0086, \u00ce\u00b8) [Token] We refer to the components on the right hand side as the lexicon, parameter, and token component respectively.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"If the key distribution\u2019s range is also known (i.e. vocabulary identifiers range from 0 to the number of words), then interpolation search can use this information instead of reading A[0] and A[|A |\u2212 1] to estimate pivots; this optimization alone led to a 24% speed improvement.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The corpus has been annotated with six different types of information, which are characterized in the following subsections.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"An inverted alignment is defined as follows: inverted alignment: i ! j = bi: Target positions i are mapped to source positions bi.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A tree set may be said to have dependencies between paths if some &quot;appropriate&quot; subset can be shown to have dependent paths as defined above.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We evaluated BABAR on two domains: terrorism and natural disasters.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The samples from each corpus were independently evaluated.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Each parse is converted into a set of constituents represented as a tuples: (label, start, end).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Parser 3, the most accurate parser, was chosen 71% of the time, and Parser 1, the least accurate parser was chosen 16% of the time.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"In contrast to these approaches, our method directly incorporates these constraints into the structure of the model.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In MUC6, the best result is achieved by SRA (Krupka, 1995).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"of Tokens No.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Besides size of training data, the use of dictionaries is another factor that might affect performance.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"This approach is similar to BABAR in that they both acquire knowledge from earlier resolutions.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"It is important to bear in mind, though, that this is not an inherent limitation of the model.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"It may be more realistic to replace the second criteria with a softer one, for example (Blum and Mitchell 98) suggest the alternative Alternatively, if Ii and 12 are probabilistic learners, it might make sense to encode the second constraint as one of minimizing some measure of the distance between the distributions given by the two learners.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"On several languages, we report performance exceeding that of more complex state-of-the art systems.1","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"This distributional sparsity of syntactic tags is not unique to English 1 The source code for the work presented in this paper is available at http:\/\/groups.csail.mit.edu\/rbg\/code\/typetagging\/.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Model Overview The model starts by generating a tag assignment T for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"For parsing, this is a mistake, especially in the case of interrogatives.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We showed in \u00c2\u00a72 that lexical ambiguity explains the underperformance of these categories.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For even larger models, we recommend RandLM; the memory consumption of the cache is not expected to grow with model size, and it has been reported to scale well.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"When two partial hypotheses have equal state (including that of other features), they can be recombined and thereafter efficiently handled as a single packed hypothesis.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The paper explains the design decisions taken in the annotations, and describes a number of applications using this corpus with its multi-layer annotation.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"( b ) s u p p o r t s c a n d i d a t e i f s e l e c t e d s e m a n t i c t a g s m a t c h t h o s e o f t h e a n a p h o r . Le xic al computes degree of lexical overlap b e t w e e n t h e c a n d i d a t e a n d t h e a n a p h o r . Re cen cy computes the relative distance between the c a n d i d a t e a n d t h e a n a p h o r . Sy nR ole computes relative frequency with which the c a n d i d a t e \u00e2\u0080\u0099 s s y n t a c t i c r o l e o c c u r s i n r e s o l u t i o n s . Figure 4: General Knowledge Sources The Lexical KS returns 1 if the candidate and anaphor are identical, 0.5 if their head nouns match, and 0 otherwise.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We characterize a class of formalisms that have this property in Section 4.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Our analysis identifies three key factors driving our performance gain: 1) selecting a model structure which directly encodes tag sparsity, 2) a type-level prior on tag assignments, and 3) a straightforward na\u00c2\u00a8\u00c4\u00b1veBayes approach to incorporate features.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Sometime, multiple words are needed, like \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, \u00e2\u0080\u009cprime minister\u00e2\u0080\u009d or \u00e2\u0080\u009cpay for\u00e2\u0080\u009d (\u00e2\u0080\u009cpay\u00e2\u0080\u009d and \u00e2\u0080\u009cpay for\u00e2\u0080\u009d are different senses in the CC-domain).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Given this similarity function, we define a nearest neighbor graph, where the edge weight for the n most similar vertices is set to the value of the similarity function and to 0 for all other vertices.","label":0},{"source_text":"There is no global pruning.","target_text":"This number must be less than or equal to n \u00f4\u0080\u0080\u0080 1.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"1 55.8 38.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"was done by the participants.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Since guess and gold trees may now have different yields, the question of evaluation is complex.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Although this is not a precise criterion, most cases we evaluated were relatively clear-cut.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"MENE has only been tested on MUC7.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"On one end of the spectrum are clustering approaches that assign a single POS tag to each word type (Schutze, 1995; Lamar et al., 2010).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"So, 1: f, xue2shengl+men0 (student+PL) 'students' occurs and we estimate its cost at 11.43; similarly we estimate the cost of f, jiang4+men0 (general+PL) 'generals' (as in 'J' f, xiao3jiang4+men0 'little generals'), at 15.02.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Otherwise, label the training data with the combined spelling\/contextual decision list, then induce a final decision list from the labeled examples where all rules (regardless of strength) are added to the decision list.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The following recursive equation is evaluated: Qe0 (e; S; C; j) = (2) = p(fj je) max \u00c3\u0086;e00 np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) max (S0;j0) (S0 ;Cnfjg;j0)!(S;C;j) j02Cnfjg Qe00 (e0; S0;C n fjg; j0)o: The search ends in the hypotheses (I; f1; ; Jg; j).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"2.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We tokenize MWUs and their POS tags; this reduces the tag set size to 12.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"However, it is desirable if we can separate them.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"\u00e2\u0080\u0094ested in correlations between prosody and dis course structure.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":".","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In this paper, we show how non-projective dependency parsing can be achieved by combining a datadriven projective parser with special graph transformation techniques.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Certainly these linguistic factors increase the difficulty of syntactic disambiguation.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Microsoft\u2019s approach uses dependency trees, others use hierarchical phrase models.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In this situation, BABAR takes the conservative approach and declines to make a resolution.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Step 4.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"However their work did not consider other types of lexical expectations (e.g., PP arguments), semantic expectations, or context comparisons like our case- frame network.(Niyu et al., 1998) used unsupervised learning to ac quire gender, number, and animacy information from resolutions produced by a statistical pronoun resolver.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Two subjects are each given a calendar and they are asked to schedule a meeting.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"1 55.8 38.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"To improve agreement during the revision process, a dual-blind evaluation was performed in which 10% of the data was annotated by independent teams.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Several systems propose statistical methods for handling unknown words (Chang et al. 1992; Lin, Chiang, and Su 1993; Peng and Chang 1993).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The complexity of the quasimonotone search is O(E3 J (R2+LR)).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The second case involves existential noun phrases (Allen, 1995), which are noun phrases that uniquely specify an object or concept and therefore do not need a prior referent in the discourse.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The core of Yarowsky's algorithm is as follows: where h is defined by the formula in equation 2, with counts restricted to training data examples that have been labeled in step 2.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"This is similar to stacking the different feature instantiations into long (sparse) vectors and computing the cosine similarity between them.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Some approaches depend upon some form of con\u00c2\u00ad straint satisfaction based on syntactic or semantic features (e.g., Yeh and Lee [1991], which uses a unification-based approach).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"computing the precision of the other's judgments relative to this standard.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We would like to thank Prof. Ralph Grish- man, Mr. Takaaki Hasegawa and Mr. Yusuke Shinyama for useful comments, discussion and evaluation.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":", for A. T.&T. nonalpha.. .","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"An extension of the TAG system was introduced by Joshi et al. (1975) and later redefined by Joshi (1987) in which the adjunction operation is defined on sets of elementary trees rather than single trees.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We checked whether the discovered links are listed in WordNet.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The parser builds dependency graphs by traversing the input from left to right, using a stack to store tokens that are not yet complete with respect to their dependents.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We refer to a segment and its assigned PoS tag as a lexeme, and so analyses are in fact sequences of lexemes.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Second, the reduced number of hidden variables and parameters dramatically speeds up learning and inference.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"t(i).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"A modified language model probability p\u00c3\u0086(eje0; e00) is defined as follows: p\u00c3\u0086(eje0; e00) =  1:0 if \u00c3\u0086 = 0 p(eje0; e00) if \u00c3\u0086 = 1 : We associate a distribution p(\u00c3\u0086) with the two cases \u00c3\u0086 = 0 and \u00c3\u0086 = 1 and set p(\u00c3\u0086 = 1) = 0:7.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The P (T |\u00cf\u0088) distribution, in English for instance, should have very low mass for the DT (determiner) tag, since determiners are a very small portion of the vocabulary.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"There has been a lot of research on such lexical relations, along with the creation of resources such as WordNet.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"41.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Dictionaries: Due to the limited amount of training material, name dictionaries have been found to be useful in the named entity task.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Roughly speaking, we say that a tee set contains trees with dependent paths if there are two paths p., = vim., and g., = in each 7 E r such that v., is some, possibly empty, shared initial subpath; v., and wi are not bounded in length; and there is some &quot;dependence&quot; (such as equal length) between the set of all v., and w., for each 7 Er.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But for eign learners are often surprised by the verbless predications that are frequently used in Arabic.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"[Hasegawa et al. 04] reported only on relation discovery, but one could easily acquire para phrases from the results.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The second weakness is purely conceptual, and probably does not affect the per\u00c2\u00ad formance of the model.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"35 76.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"(1999) did not report using any dictionaries, but mentioned in a footnote that they have added list membership features, which have helped marginally in certain domains.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This model is easily incorporated into the segmenter by building a WFST restrict\u00c2\u00ad ing the names to the four licit types, with costs on the arcs for any particular name summing to an estimate of the cost of that name.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Nonetheless there is no alternative to Reiche\u00e2\u0080\u0099s plan.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"We apply a beam search concept as in speech recognition.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In this section, we describe the seven general knowledge sources and explain how the DempsterShafer model makes resolutions.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Our analysis identifies three key factors driving our performance gain: 1) selecting a model structure which directly encodes tag sparsity, 2) a type-level prior on tag assignments, and 3) a straightforward na\u00c2\u00a8\u00c4\u00b1veBayes approach to incorporate features.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"As each global feature group is added to the list of features, we see improvements to both MUC6 and MUC6 MUC7 Baseline 90.75% 85.22% + ICOC 91.50% 86.24% + CSPP 92.89% 86.96% + ACRO 93.04% 86.99% + SOIC 93.25% 87.22% + UNIQ 93.27% 87.24% Table 3: F-measure after successive addition of each global feature group Table 5: Comparison of results for MUC6 Systems MUC6 MUC7 No.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"we perform five runs with different random initialization of sampling state.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, both a chair and a truck would be labeled as artifacts, but this does not at all suggest that they are coreferent.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"orthographic words are thus only a starting point for further analysis and can only be regarded as a useful hint at the desired division of the sentence into words.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We tried two versions of our graph-based approach: feature after the first stage of label propagation (Eq.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Step 2.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Phrase-level granularity distinguishes our work from previous work by Matsoukas et al (2009), who weight sentences according to sub-corpus and genre membership.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"To bridge this gap, we consider a practically motivated scenario, in which we want to leverage existing resources from a resource-rich language (like English) when building tools for resource-poor foreign languages.1 We assume that absolutely no labeled training data is available for the foreign language of interest, but that we have access to parallel data with a resource-rich language.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"3.2 Stochastic rhetorical analysis.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The results in Table 1 show PROBING is 81% faster than TRIE, which is in turn 31% faster than the fastest baseline.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The pre terminal morphological analyses are mapped to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags provided with the tree- bank.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For each pair of judges, consider one judge as the standard,.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The aforementioned surface form bcl, for example, may also stand for the lexical item \u201conion\u201d, a Noun.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We also thank ChaoHuang Chang, reviewers for the 1994 ACL conference, and four anonymous reviewers for Computational Linguistics for useful comments.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"As we have seen, the lexicon of basic words and stems is represented as a WFST; most arcs in this WFST represent mappings between hanzi and pronunciations, and are costless.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"And if one is interested in TIS, one would probably consider the single orthographic word ACL to consist of three phonological words-lei s'i d\/-corresponding to the pronunciation of each of the letters in the acronym.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Reading the following record\u2019s offset indicates where the block ends.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The development of the very first Hebrew Treebank (Sima\u2019an et al., 2001) called for the exploration of general statistical parsing methods, but the application was at first limited.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In the present study, we limit ourselves to an algorithmic approach, using a deterministic breadthfirst search.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The most popular approach to dealing with seg\u00c2\u00ad mentation ambiguities is the maximum matching method, possibly augmented with further heuristics.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"A spelling rule might be a simple look-up for the string (e.g., a rule that Honduras is a location) or a rule that looks at words within a string (e.g., a rule that any string containing Mr. is a person).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Figure 2 shows examples of extracted NE pair instances and their contexts.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"8 Volume 1, English language, 19961997, Format version 1, correction level 0 An ap ho r T e r r o r i s m R e c Pr F D i s a s t e r s R e c Pr F De f. NP s Pro no uns .43 .79 .55 .50 .72 .59 .42 .91 .58 .42 .82 .56 Tot al .46 .76 .57 .42 .87 .57 Table 2: General Knowledge Sources Table 4: Individual Performance of KSs for Terrorism Table 3: General + Contextual Role Knowledge Sources larger MUC4 and Reuters corpora.9 4.2 Experiments.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"English parsing evaluations usually report results on sentences up to length 40.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For each terminal, the Leaf Ancestor metric extracts the shortest path to the root.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In this situation, BABAR takes the conservative approach and declines to make a resolution.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"2.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"But Arabic contains a variety of linguistic phenomena unseen in English.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The transitive closure of the dictionary in (a) is composed with Id(input) (b) to form the WFST (c).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Purely statistical approaches have not been very popular, and so far as we are aware earlier work by Sproat and Shih (1990) is the only published instance of such an approach.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Hence, there is a total of 29 classes (7 name classes 4 sub-classes 1 not-a-name class).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"2) An improved language model, which takes into account syntactic structure, e.g. to ensure that a proper English verbgroup is generated.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"We have presented a new method for non-projective dependency parsing, based on a combination of data-driven projective dependency parsing and graph transformation techniques.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"As will be obvious later, their derivation tree sets will be local sets as are those of CFG's.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Since our objective is to compare distributions of bracketing discrepancies, we do not use heuristics to prune the set of nuclei.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We can better predict the probability of an unseen hanzi occurring in a name by computing a within-class Good-Turing estimate for each radical class.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In principle a feature could be an arbitrary predicate of the (spelling, context) pair; for reasons that will become clear, features are limited to querying either the spelling or context alone.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"As a result, CFG's can not provide the structural descriptions in which there are nested dependencies between symbols labelling a path.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"What is important and is not expressed by the notation is the so-called coverage constraint: each source position j should be 'hit' exactly once by the path of the inverted alignment bI 1 = b1:::bi:::bI . Using the inverted alignments in the maximum approximation, we obtain as search criterion: max I (p(JjI) max eI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) max bI 1 I Yi=1 [p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])) = = max I (p(JjI) max eI 1;bI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])); where the two products over i have been merged into a single product over i. p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) is the trigram language model probability.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Using structural information As was explained in the results section, we extracted examples like \u00e2\u0080\u009cSmith estimates Lotus\u00e2\u0080\u009d, from a sentence like \u00e2\u0080\u009cMr.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Section 4 draws some conclusions from the present state of the effort.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The role that each noun phrase plays in the kidnapping event is key to distinguishing these cases.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"An inverted alignment is defined as follows: inverted alignment: i ! j = bi: Target positions i are mapped to source positions bi.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Suffixes and Prefixes: This group contains only two features: Corporate-Suffix and Person-Prefix.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Our second point of comparison is with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For example Chen and Liu (1992) report precision and recall rates of over 99%, but this counts only the words that occur in the test corpus that also occur in their dictionary.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"One class comprises words derived by productive morphologi\u00c2\u00ad cal processes, such as plural noun formation using the suffix ir, menD.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"An edge indicates that the two features must have the same label.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"It is unique, agrees with the maximum-likelihood distribution, and has the exponential form (Della Pietra et al., 1997): where refers to the outcome, the history (or context), and is a normalization function.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"However, it is also mostly political content (even if not focused on the internal workings of the European Union) and opinion.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The baseline simply retains the original labels for all arcs, regardless of whether they have been lifted or not, and the number of distinct labels is therefore simply the number n of distinct dependency types.2 In the first encoding scheme, called Head, we use a new label d\u2191h for each lifted arc, where d is the dependency relation between the syntactic head and the dependent in the non-projective representation, and h is the dependency relation that the syntactic head has to its own head in the underlying structure.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"One may argue with these efforts on normalization, and ultimately their value should be assessed by assessing their impact on inter-annotator agreement.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We train linear mixture models for conditional phrase pair probabilities over IN and OUT so as to maximize the likelihood of an empirical joint phrase-pair distribution extracted from a development set.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Borth- wick (1999) successfully made use of other hand- coded systems as input for his MENE system, and achieved excellent results.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This paper proposes a simple and effective tagging method that directly models tag sparsity and other distributional properties of valid POS tag assignments.","label":0},{"source_text":"There is no global pruning.","target_text":"For our demonstration system, we typically use the pruning threshold t0 = 5:0 to speed up the search by a factor 5 while allowing for a small degradation in translation accuracy.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Notice that the CC-domain is a special case.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In these examples, the names identified by the two systems (if any) are underlined; the sentence with the correct segmentation is boxed.19 The differences in performance between the two systems relate directly to three issues, which can be seen as differences in the tuning of the models, rather than repre\u00c2\u00ad senting differences in the capabilities of the model per se.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"If we were working with more than three parsers we could investigate minority constituents, those constituents that are suggested by at least one parser, but which the majority of the parsers do not suggest.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In fact, training on the official training data is not suitable as the articles in this data set are entirely about aviation disasters, and the test data is about air vehicle launching.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Mikheev et al.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Our model outperforms theirs on four out of five languages on the best hyperparameter setting and three out of five on the median setting, yielding an average absolute difference across languages of 12.9% and 3.9% for best and median settings respectively compared to their best EM or LBFGS performance.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"One key issue here is to seek a discourse-based model of information structure.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"42 nator, the N31s can be measured well by counting, and we replace the expectation by the observation.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"This is the form of recursive levels in iDafa constructs.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The iw all map variant uses a non-0 y weight on a uniform prior in p,,(s t), and outperforms a version with y = 0 (iw all) and the \u201cflattened\u201d variant described in section 3.2.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"4.1 Corpora.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Conversely, the lattice parser requires no linguistic resources and produces segmentations of comparable quality.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"1 2 3.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"(Hartmann 1984), for example, used the term Reliefgebung to characterize the distibution of main and minor information in texts (similar to the notion of nuclearity in RST).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"(S2) The burglar fired the gun three times and fled.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Generalizing state minimization, the model could also provide explicit bounds on probability for both backward and forward extension.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The availability of comparable corpora is limited, which is a significant limitation on the approach.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The taggers were trained on datasets labeled with the universal tags.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In Section 3, we introduce our novel concept to word reordering and a DP-based search, which is especially suitable for the translation direction from German to English.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The same form fmnh can be segmented as f-mnh, f (\u201cthat\u201d) functioning as a reletivizer with the form mnh.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"How should the absence of vowels and syntactic markers influence annotation choices and grammar development?","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The global feature groups are: InitCaps of Other Occurrences (ICOC): There are 2 features in this group, checking for whether the first occurrence of the same word in an unambiguous position (non first-words in the TXT or TEXT zones) in the same document is initCaps or not-initCaps.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"(c) After they blindfolded the men...","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Limitations There are several limitations in the methods.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"6) are noisy, the results confirm that label propagation within the foreign language part of the graph adds significant quality for every language.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The contextual role knowledge that BABAR uses for coreference resolution is derived from this caseframe data.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"An easy way to achieve this is to put the domain-specific LMs and TMs into the top-level log-linear model and learn optimal weights with MERT (Och, 2003).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"As discussed in more detail in \u00a73, we use two types of vertices in our graph: on the foreign language side vertices correspond to trigram types, while the vertices on the English side are individual word types.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"As illustrated by MCTAG's, it is possible for a formalism to give tree sets with bounded dependent paths while still sharing the constrained rewriting properties of CFG's, HG's, and TAG's.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In order to create an IE system for a new domain, one has to spend a long time to create the knowledge.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"2 62.2 39.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Initially, the DempsterShafer model assumes that all hypotheses are equally likely, so it creates a set called \u00ce\u00b8 that includes all hypotheses.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"TPT Germann et al. (2009) describe tries with better locality properties, but did not release code.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We first define &quot;pseudo-labels&quot;,-yt, as follows: = Yi t sign(g 0\\ 2\u2014 kx2,m < i < n Thus the first m labels are simply copied from the labeled examples, while the remaining (n \u2014 m) examples are taken as the current output of the second classifier.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Skipped (K): The translation of up to one word may be postponed . Verb (V): The translation of up to two words may be anticipated.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Unique Occurrences and Zone (UNIQ): This group of features indicates whether the word is unique in the whole document.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Evaluation within a set The evaluation of paraphrases within a set of phrases which share a keyword is illustrated in Figure 4.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"From now on all lattice arcs are tagged segments and the assignment of probability P(p \u2014* (s, p)) to lattice arcs proceeds as usual.4 A rather pathological case is when our lexical heuristics prune away all segmentation possibilities and we remain with an empty lattice.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Some relations are signalled by subordinating conjunctions, which clearly demarcate the range of the text spans related (matrix clause, embedded clause).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In our model, we associate these features at the type-level in the lexicon.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"RandLM 0.2 (Talbot and Osborne, 2007) stores large-scale models in less memory using randomized data structures.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The goal of machine translation is the translation of a text given in some source language into a target language.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"It falls short of the \u201cProjection\u201d baseline for German, but is statistically indistinguishable in terms of accuracy.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"However, it is also mostly political content (even if not focused on the internal workings of the European Union) and opinion.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We use the log-linear tagger of Toutanova et al.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Unigram records store probability, backoff, and an index in the bigram table.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"7).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"To investigate the influence of these factors, we analyze Modern Standard Arabic (henceforth MSA, or simply \u00e2\u0080\u009cArabic\u00e2\u0080\u009d) because of the unusual opportunity it presents for comparison to English parsing results.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"For all languages we do not make use of a tagging dictionary.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Subsets of partial hypotheses with coverage sets C of increasing cardinality c are processed.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The first 3770 trees of the resulting set then were used for training, and the last 418 are used testing.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The baseline alignment model does not permit that a source word is aligned to two or more target words, e.g. for the translation direction from German toEnglish, the German compound noun 'Zahnarztter min' causes problems, because it must be translated by the two target words dentist's appointment.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective treebanks is often neglected because of the relative scarcity of problematic constructions.","label":1},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The TRIE model continues to use the least memory of ing (-P) with MAP POPULATE, the default.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The first m pairs have labels yi, whereas for i = m + 1, , n the pairs are unlabeled.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We have developed a general approach for combining parsers when preserving the entire structure of a parse tree is important.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Finally, we wish to reiterate an important point.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"This left 962 examples, of which 85 were noise.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For all lists except locations, the lists are processed into a list of tokens (unigrams).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The ATB has a much higher fraction of nuclei per tree, and a higher type-level error rate.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The second model (+PRIOR) utilizes the independent prior over type-level tag assignments P (T |\u00cf\u0088).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The final strong hypothesis, denoted 1(x), is then the sign of a weighted sum of the weak hypotheses, 1(x) = sign (Vii atht(x)), where the weights at are determined during the run of the algorithm, as we describe below.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Figure 2: An ATB sample from the human evaluation.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The learned information was recycled back into the resolver to improve its performance.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This modification brings the method closer to the DL-CoTrain algorithm described earlier, and is motivated by the intuition that all three labels should be kept healthily populated in the unlabeled examples, preventing one label from dominating \u2014 this deserves more theoretical investigation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The model we use provides a simple framework in which to incorporate a wide variety of lexical information in a uniform way.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Second, the reduced number of hidden variables and parameters dramatically speeds up learning and inference.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"This offers the well-known advantages for inter- changability, but it raises the question of how to query the corpus across levels of annotation.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Nodes in the trie are based on arrays sorted by vocabulary identifier.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Only 2 link in the CC- domain (buy-purchase, acquire-acquisition) and 2 links (trader-dealer and head-chief) in the PC- domain are found in the same synset of Word- Net 2.1 (http:\/\/wordnet.princeton.edu\/).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The POS distributions over the foreign trigram types are used as features to learn a better unsupervised POS tagger (\u00a75).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In addition, this formulation results in a dramatic reduction in the number of model parameters thereby, enabling unusually rapid training.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Here we do not submit to this view.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For t = 1, T and for j = 1, 2: where 4 = exp(-jg'(xj,i)). practice, this greedy approach almost always results in an overall decrease in the value of Zco.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Coverage indicates the fraction of hypotheses in which the character yield exactly matched the reference.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Both of these analyses are shown in Figure 4; fortunately, the correct analysis is also the one with the lowest cost, so it is this analysis that is chosen.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We substantially outperform all of them on query speed and offer lower memory consumption than lossless alternatives.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In this work, we take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"For the PROBING implementation, hash table sizes are in the millions, so the most relevant values are on the right size of the graph, where linear probing wins.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Cohen and Smith approach this by introducing the \u03b1 hyperparameter, which performs best when optimized independently for each sentence (cf.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Supervised part-of-speech (POS) taggers, for example, approach the level of inter-annotator agreement (Shen et al., 2007, 97.3% accuracy for English).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We also thank ChaoHuang Chang, reviewers for the 1994 ACL conference, and four anonymous reviewers for Computational Linguistics for useful comments.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"evaluated to account for the same fraction of the data.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Once HMM parameters (\u00ce\u00b8, \u00cf\u0086) are drawn, a token-level tag and word sequence, (t, w), is generated in the standard HMM fashion: a tag sequence t is generated from \u00cf\u0086.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"2.2.3 Lexical Caseframe Expectations The second type of contextual role knowledge learned by BABAR is Lexical Caseframe Expectations, which are used by the CFLex knowledge source.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"As the reviewer also points out, this is a problem that is shared by, e.g., probabilistic context-free parsers, which tend to pick trees with fewer nodes.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Even without features, but still using the tag prior, our median result is 52.0%, still significantly outperforming Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Previous Work.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"(c) After they blindfolded the men...","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"SSER: subjective sentence error rate: For a more detailed analysis, the translations are judged by a human test person.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The developers aimed to reduce memory consumption at the expense of time.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Lexicon and OOV Handling Our data-driven morphological-analyzer proposes analyses for unknown tokens as described in Section 5.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"2.3 Rhetorical structure.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Log-linear combination (loglin) improves on this in all cases, and also beats the pure IN system.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Our clue is the NE instance pairs.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, the passive voice pattern \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d and the active voice pattern \u00e2\u0080\u009ckidnapped <direct object>\u00e2\u0080\u009d are merged into a single normalized pattern \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d.2 For the sake of sim plicity, we will refer to these normalized extraction patterns as caseframes.3 These caseframes can capture two types of contextual role information: (1) thematic roles corresponding to events (e.g, \u00e2\u0080\u009c<agent> kidnapped\u00e2\u0080\u009d or \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d), and (2) predicate-argument relations associated with both verbs and nouns (e.g., \u00e2\u0080\u009ckidnapped for <np>\u00e2\u0080\u009d or \u00e2\u0080\u009cvehicle with <np>\u00e2\u0080\u009d).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"It is worth noting that the middle words of the Italian trigrams are nouns too, which exhibits the fact that the similarity metric connects types having the same syntactic category.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We did have a problem classifying some modified noun phrases where the modified phrase does not represent a qualified or restricted form of the head, like \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, as these are both represented by the keyword \u00e2\u0080\u009cchairman\u00e2\u0080\u009d.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"They also describe an application of cotraining to classifying web pages (the to feature sets are the words on the page, and other pages pointing to the page).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"One hopes that such a corpus will be forth\u00c2\u00ad coming.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"1 1 0.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The first modification \u2014 cautiousness \u2014 is a relatively minor change.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"On each step CoBoost searches for a feature and a weight so as to minimize either 40 or 40.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The equation for sampling a single type-level assignment Ti is given by, 0.2 0 5 10 15 20 25 30 Iteration Figure 2: Graph of the one-to-one accuracy of our full model (+FEATS) under the best hyperparameter setting by iteration (see Section 5).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, this result is consistent with the results of ex\u00c2\u00ad periments discussed in Wu and Fung (1994).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The token-level term is similar to the standard HMM sampling equations found in Johnson (2007).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Interpolation search has a more expensive pivot but performs less pivoting and reads, so it is slow on small data and faster on large data.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For example, if a token starts with a capital letter and ends with a period (such as Mr.), then the feature InitCapPeriod is set to 1, etc. First Word: This feature group contains only one feature firstword.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"When this metric is less than 0.5, we expect to incur more errors' than we will remove by adding those constituents to the parse.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"First we present the non-parametric version of parser switching, similarity switching: The intuition for this technique is that we can measure a similarity between parses by counting the constituents they have in common.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"9 66.4 47.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(Blum and Mitchell 98) go on to give PAC results for learning in the cotraining case.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"4.1 Local Features.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We realize the importance of paraphrase; however, the major obstacle is the construction of paraphrase knowledge.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The choice of the particular newspaper was motivated by the fact that the language used in a regional daily is somewhat simpler than that of papers read nationwide.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Instead, we resort to an iterative update based method.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"As indicated in Figure 1(c), apart from this correct analysis, there is also the analysis taking B ri4 as a word (e.g., a common abbreviation for Japan), along with X:\u00c2\u00a5 wen2zhangl 'essay\/ and f!!.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"attaching to terms denoting human beings.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"These models generally outperform our memory consumption but are much slower, even when cached.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"In both cases, the instanceweighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline, and gains of between 0.6 and 1.8 over an equivalent mixture model (with an identical training procedure but without instance weighting).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For example, the independence assumptions mean that the model fails to capture the dependence between specific and more general features (for example the fact that the feature full.-string=New_York is always seen with the features contains (New) and The baseline method tags all entities as the most frequent class type (organization). contains (York) and is never seen with a feature such as contains (Group) ).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Crucially, the conventional orthographic form of MSA text is unvocalized, a property that results in a deficient graphical representation.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The parallel data came from the Europarl corpus (Koehn, 2005) and the ODS United Nations dataset (UN, 2006).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We plan to explore more powerful techniques for exploiting the diversity of parsing methods.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Still, for both human and automatic rhetorical analysis, connectives are the most important source of surface information.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Note, however, that there might be situations in which Zco in fact increases.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This variant is tested in the experiments below.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In defining LCFRS's, we hope to generalize the definition of CFG's to formalisms manipulating any structure, e.g. strings, trees, or graphs.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"NN \u00ef\u00bf\u00bd .e NP NNP NP DTNNP NN \u00ef\u00bf\u00bd .e NP NP NNP NP Table 5: Evaluation of 100 randomly sampled variation nuclei types.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"of Tokens No.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"A similar structure is used in speech recognition.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In our work, we demonstrate that using a simple na\u00c2\u00a8\u00c4\u00b1veBayes approach also yields substantial performance gains, without the associated training complexity.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Each out-of-domain phrase pair is characterized by a set of simple features intended to reflect how useful it will be.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"18 We are grateful to ChaoHuang Chang for providing us with this set.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We conjecture that this trend may continue by incorporating additional information, e.g., three-dimensional models as proposed by Tsarfaty and Sima\u2019an (2007).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(5) and ht into Equ.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"If the context wnf will never extend to the right (i.e. wnf v is not present in the model for all words v) then no subsequent query will match the full context.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"As lower frequency examples include noise, we set a threshold that an NE category pair should appear at least 5 times to be considered and an NE instance pair should appear at least twice to be considered.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Although these authors report better gains than ours, they are with respect to a non-adapted baseline.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The points enumerated above are particularly related to ITS, but analogous arguments can easily be given for other applications; see for example Wu and Tseng's (1993) discussion of the role of segmentation in information retrieval.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The perplexity for the trigram language model used is 26:5.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"2.2.1 The Caseframe Representation Information extraction (IE) systems use extraction patterns to identify noun phrases that play a specific role in 1 Our implementation only resolves NPs that occur in the same document, but in retrospect, one could probably resolve instances of the same existential NP in different documents too.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"For instance, by altering the emission distribution parameters, Johnson (2007) encourages the model to put most of the probability mass on few tags.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Therefore, the number of fine tags varied across languages for our experiments; however, one could as well have fixed the set of HMM states to be a constant across languages, and created one mapping to the universal POS tagset.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Search CPU time mWER SSER Method [sec] [%] [%] MonS 0:9 42:0 30:5 QmS 10:6 34:4 23:8 IbmS 28:6 38:2 26:2 4.2 Performance Measures.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"annotation guidelines that tell annotators what to do in case of doubt.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Our new approach, which is called quasi-monotone search, processes the source sentence monotonically, while explicitly taking into account the positions of the German verbgroup.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We call such a constituent an isolated constituent.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In the second and third translation examples, the IbmS word reordering performs worse than the QmS word reordering, since it can not take properly into account the word reordering due to the German verbgroup.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(1992).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"To check for this, we do pairwise bootstrap resampling: Again, we repeatedly sample sets of sentences, this time from both systems, and compare their BLEU scores on these sets.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For that application, at a minimum, one would want to know the phonological word boundaries.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Figure 3 Figure 1.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Furthermore, we know one of the original parses will be the hypothesized parse, so the direct method of determining which one is best is to compute the probability of each of the candidate parses using the probabilistic model we developed in Section 2.1.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Our monolingual similarity function (for connecting pairs of foreign trigram types) is the same as the one used by Subramanya et al. (2010).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We will report the evaluation results in the next subsection.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"attaching to terms denoting human beings.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Church and Hanks [1989]), and we have used lists of character pairs ranked by mutual information to expand our own dictionary.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The high 1 tone of J1l would not normally neutralize in this fashion if it were functioning as a word on its own.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"On the English side, however, the vertices (denoted by Ve) correspond to word types.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Unfortunately, there is no standard corpus of Chinese texts, tagged with either single or multiple human judgments, with which one can compare performance of various methods.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Specifically, we (+FEATS) utilizes the tag prior as well as features (e.g., suffixes and orthographic features), discussed in Section 3, for the P (W |T , \u00cf\u0088) component.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In TAG's the elementary tree and addresses where adjunction takes place are used to instantiate the operation.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Cohen and Smith (2007) chose a metric like SParseval (Roark et al., 2006) that first aligns the trees and then penalizes segmentation errors with an edit-distance metric.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Both parametric and non-parametric models are explored.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"BABAR uses the log-likelihood statistic (Dunning, 1993) to evaluate the strength of a co-occurrence relationship.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Many statistical or machine-learning approaches for natural language problems require a relatively large amount of supervision, in the form of labeled training examples.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Each feature group can be made up of many binary features.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Some relations are signalled by subordinating conjunctions, which clearly demarcate the range of the text spans related (matrix clause, embedded clause).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"As was explained in the results section, \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d are not desirable keywords in the CC-domain.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Can we do . QmS: Yes, wonderful.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"This approach is compared to another reordering scheme presented in (Berger et al., 1996).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Fourth, we show how to build better models for three different parsers.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Dynamic programming efficiently scores many hypotheses by exploiting the fact that an N-gram language model conditions on at most N \u2212 1 preceding words.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"In Table 1 we see with very few exceptions that the isolated constituent precision is less than 0.5 when we use the constituent label as a feature.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"IRSTLM (Federico et al., 2008) is an open-source toolkit for building and querying language models.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Note that while the standard HMM, has O(K n) emission parameters, our model has O(n) effective parameters.3 Token Component Once HMM parameters (\u00cf\u0086, \u00ce\u00b8) have been drawn, the HMM generates a token-level corpus w in the standard way: P (w, t|\u00cf\u0086, \u00ce\u00b8) = P (T , W , \u00ce\u00b8, \u00cf\u0088, \u00cf\u0086, t, w|\u00ce\u00b1, \u00ce\u00b2) = P (T , W , \u00cf\u0088|\u00ce\u00b2) [Lexicon] \u00ef\u00a3\u00ab n n \u00ef\u00a3\u00ad (w,t)\u00e2\u0088\u0088(w,t) j \u00ef\u00a3\u00b6 P (tj |\u00cf\u0086tj\u00e2\u0088\u00921 )P (wj |tj , \u00ce\u00b8tj )\u00ef\u00a3\u00b8 P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) [Parameter] P (w, t|\u00cf\u0086, \u00ce\u00b8) [Token] We refer to the components on the right hand side as the lexicon, parameter, and token component respectively.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The set of n-grams appearing in a model is sparse, and we want to efficiently find their associated probabilities and backoff penalties.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We believe that the underlying principles of the maximum entropy framework are suitable for exploiting information from diverse sources.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"To show that the derivation tree set of a TAG is a local set, nodes are labeled by pairs consisting of the name of an elementary tree and the address at which it was adjoined, instead of labelling edges with addresses.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The state-to-tag mapping is obtained from the best hyperparameter setting for 11 mapping shown in Table 3.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"However, those methods need initial seeds, so the relation between entities has to be known in advance.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Figure 4 shows the seven general knowledge sources (KSs) that represent features commonly used for coreference resolution.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The above auxiliary quantity satisfies the following recursive DP equation: Qe0 (e; C; j) = Initial Skip Verb Final 1.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Before presenting our results, we describe the datasets that we used, as well as two baselines.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(a) ;IE shi4 'be' => ;IE;IE shi4bu2-shi4 (be-not-be) 'is it?'","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Specifically, we (+FEATS) utilizes the tag prior as well as features (e.g., suffixes and orthographic features), discussed in Section 3, for the P (W |T , \u00cf\u0088) component.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Word Head Of Complement POS 1 '01 inna \u00e2\u0080\u009cIndeed, truly\u00e2\u0080\u009d VP Noun VBP 2 '01 anna \u00e2\u0080\u009cThat\u00e2\u0080\u009d SBAR Noun IN 3 01 in \u00e2\u0080\u009cIf\u00e2\u0080\u009d SBAR Verb IN 4 01 an \u00e2\u0080\u009cto\u00e2\u0080\u009d SBAR Verb IN Table 1: Diacritized particles and pseudo-verbs that, after orthographic normalization, have the equivalent surface form 0 an.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Moses sets the cache size parameter to 50 so we did as well; the resulting cache size is 2.82 GB.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We will report the evaluation results in the next subsection.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Further, the probing hash table does only one random lookup per query, explaining why it is faster on large data.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"the number of permutations carried out for the word reordering is given.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Since our goal is to perform well under these measures we will similarly treat constituents as the minimal substructures for combination.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"It is formally straightforward to extend the grammar to include these names, though it does increase the likelihood of overgeneration and we are unaware of any working systems that incorporate this type of name.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Thus it is possible, for illustration, to look for a noun phrase (syntax tier) marked as topic (information structure tier) that is in a bridging relation (co-reference tier) to some other noun phrase.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Evaluating parsing results in our joint framework, as argued by Tsarfaty (2006), is not trivial under the joint disambiguation task, as the hypothesized yield need not coincide with the correct one.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"One of the difficulties in Natural Language Processing is the fact that there are many ways to express the same thing or event.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Parameter Component As in the standard Bayesian HMM (Goldwater and Griffiths, 2007), all distributions are independently drawn from symmetric Dirichlet distributions: 2 Note that t and w denote tag and word sequences respectively, rather than individual tokens or tags.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The Bikel GoldPOS configuration only supplies the gold POS tags; it does not force the parser to use them.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"As an indication, in our core corpus, we found an average sentence length of 15.8 words and 1.8 verbs per sentence, whereas a randomly taken sample of ten commentaries from the national papers Su\u00c2\u00a8ddeutsche Zeitung and Frankfurter Allgemeine has 19.6 words and 2.1 verbs per sentence.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In order to handle the necessary word reordering as an optimization problem within our dynamic programming approach, we describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming (Held, Karp, 1962).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Each derivation of a grammar can be represented by a generalized context-free derivation tree.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The number of ATB n-grams also falls below the WSJ sample size as the largest WSJ sample appeared in only 162 corpus positions.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Base NPs are the other significant category of nominal phrases.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"(2009).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Then, token- level HMM emission parameters are drawn conditioned on these assignments such that each word is only allowed probability mass on a single assigned tag.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We weight edges using a unigram language model estimated with Good- Turing smoothing.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In previous work (Bean and Riloff, 1999), we developed an unsupervised learning algorithm that automatically recognizes definite NPs that are existential without syntactic modification because their meaning is universally understood.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We first give a brief overview of boosting algorithms.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"First, we identify sources of syntactic ambiguity understudied in the existing parsing literature.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"0 . 8 3 1 0.859 496 76.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The set is then compared with the set generated from the Penn Treebank parse to determine the precision and recall.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"As one can see, only the trigrams [suo incarceramento ,], [suo iter ,] and [suo carattere ,] are connected to English words.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Link phrases based on instance pairs Using NE instance pairs as a clue, we find links between sets of phrases.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The first concerns how to deal with ambiguities in segmentation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Since the transducers are built from human-readable descriptions using a lexical toolkit (Sproat 1995), the system is easily maintained and extended.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"First, we evaluated BABAR using only the seven general knowledge sources.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"For example, let us consider a tree set containing trees of the form shown in Figure 4a.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The larger sets are more accurate than the small sets.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Even with vocalization, there are linguistic categories that are difficult to identify without semantic clues.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In the last example, the less restrictive IbmS word reordering leads to a better translation, although the QmS translation is still acceptable.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Let us consider an example of ambiguity caused by devocalization.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"As noted in Section 4.4, disk cache state is controlled by reading the entire binary file before each test begins.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"All features were conjoined with the state z.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The errors shown are from the Berkeley parser output, but they are representative of the other two parsing models.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"We will show that by giving the first model some global features, MENERGI outperforms Borthwick' s reference resolution classifier.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We have presented two general approaches to studying parser combination: parser switching and parse hybridization.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The corpora for both settings are summarized in table 1.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This essentially means that their model does not result in a true joint inference, as executions for different tasks involve tuning a parameter separately.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Whereas Arabic linguistic theory as Saddam (a) Reference (b) Stanford signs (1) and (2) to the class of pseudo verbs 01 +i J>1\u00ef\u00bf\u00bd inna and her sisters since they can beinflected, the ATB conventions treat (2) as a com plementizer, which means that it must be the head of SBAR.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"P St = n. \u00ce\u00b2 T VARIABLES \u00cf\u0088 Y W : Word types (W1 ,.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For each caseframe, BABAR collects the semantic classes associated with the head nouns of NPs that were extracted by the caseframe.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This representation gives ir, an appropriate morphological decomposition, pre\u00c2\u00ad serving information that would be lost by simply listing ir, as an unanalyzed form.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"D o m ai n Li n k ac cu ra cy W N c o v e r a g e C C 7 3 . 3 % 2 \/ 1 1 P C 8 8 . 9 % 2 \/ 8 Table 2.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"A compatible view is presented by Charniak et al. (1996) who consider the kind of probabilities a generative parser should get from a PoS tagger, and concludes that these should be P(w|t) \u201cand nothing fancier\u201d.3 In our setting, therefore, the Lattice is not used to induce a probability distribution on a linear context, but rather, it is used as a common-denominator of state-indexation of all segmentations possibilities of a surface form.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In approaching this problem, a variety of different methods are conceivable, including a more or less sophisticated use of machine learning.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"\u00e2\u0080\u0094 similar results have been observed across multiple languages.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For the examples given in (1) and (2) this certainly seems possible.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The likelihood of the observed data under the model is where P(yi, xi) is defined as in (9).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"This representation of structures by substrings and the composition operation by its effect on substrings is related to the work of Rounds (1985).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The best analysis of the corpus is taken to be the true analysis, the frequencies are re-estimated, and the algorithm is repeated until it converges.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"For the error counts, a range from 0:0 to 1:0 is used.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Thus we opted not to take the step of creating more precise written annotation guidelines (as (Carlson, Marcu 2001) did for English), which would then allow for measuring inter-annotator agreement.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Rather we believe several methods have to be developed using different heuristics to discover wider variety of paraphrases.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Each treelet (an internal node with all its children) represents the use of a rule that is encapsulated by the grammar The grammar encapsulates (either explicitly or implicitly) a finite number of rules that can be written as follows: n > 0 In the case of CFG's, for each production In the case of TAG's, a derivation step in which the derived trees RI, \u2022 \u2022 \u2022 , On are adjoined into fi at rhe addresses \u2022 \u2022 \u2022 \u2022 in. would involve the use of the following rule2.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"It is well-known that constituency parsing models designed for English often do not generalize easily to other languages and treebanks.1 Explanations for this phenomenon have included the relative informativeness of lexicalization (Dubey and Keller, 2003; Arun and Keller, 2005), insensitivity to morphology (Cowan and Collins, 2005; Tsarfaty and Sima\u00e2\u0080\u0099an, 2008), and the effect of variable word order (Collins et al., 1999).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"(b) After they were released...","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"For the \u00e2\u0080\u0098core\u00e2\u0080\u0099 portion of PCC, we found that on average, 35% of the coherence relations in our RST annotations are explicitly signalled by a lexical connective.6 When adding the fact that connectives are often ambiguous, one has to conclude that prospects for an automatic analysis of rhetorical structure using shallow methods (i.e., relying largely on connectives) are not bright \u00e2\u0080\u0094 but see Sections 3.2 and 3.3 below.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"com t 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Figure 2: Order in which source positions are visited for the example given in Fig.1.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"And time is short.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Label propagation is used to propagate these tags inwards and results in tag distributions for the middle word of each Italian trigram.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Papers that use this method or minor variants thereof include Liang (1986), Li et al.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In this paper, we describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"ments contained 322 anaphoric links.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Intuitively, as suggested by the example in the introduction, this is the right granularity to capture domain effects.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"First, we describe how the caseframes are represented and learned.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This is orthographically represented as 7C.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"This formulation of the constraint feature is equivalent to the use of a tagging dictionary extracted from the graph using a threshold T on the posterior distribution of tags for a given word type (Eq.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Our results suggest that current parsing models would benefit from better annotation consistency and enriched annotation in certain syntactic configurations.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A non-optimal analysis is shown with dotted lines in the bottom frame.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.2.1 The Caseframe Representation Information extraction (IE) systems use extraction patterns to identify noun phrases that play a specific role in 1 Our implementation only resolves NPs that occur in the same document, but in retrospect, one could probably resolve instances of the same existential NP in different documents too.","label":1},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This combination generalizes (2) and (3): we use either at = a to obtain a fixed-weight linear combination, or at = cI(t)\/(cI(t) + 0) to obtain a MAP combination.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"It was filtered to retain the top 30 translations for each source phrase using the TM part of the current log-linear model.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We also report word type level accuracy, the fraction of word types assigned their majority tag (where the mapping between model state and tag is determined by greedy one-to-one mapping discussed above).5 For each language, we aggregate results in the following way: First, for each hyperparameter setting, evaluate three variants: The first model (1TW) only 4 Typically, the performance stabilizes after only 10 itera-.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"By design, they readily capture regularities at the token-level.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Previous Work.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Two sets of examples from Gan are given in (1) and (2) (:::::: Gan's Appendix B, exx.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The second case involves existential noun phrases (Allen, 1995), which are noun phrases that uniquely specify an object or concept and therefore do not need a prior referent in the discourse.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Following Dickinson (2005), we randomly sampled 100 variation nuclei from each corpus and evaluated each sample for the presence of an annotation error.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Another approach to finding paraphrases is to find phrases which take similar subjects and objects in large corpora by using mutual information of word distribution [Lin and Pantel 01].","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"A formalism generating tree sets with complex path sets can still generate only semilinear languages if its tree sets have independent paths, and semilinear path sets.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"A morpheme, on the other hand, usually corresponds to a unique hanzi, though there are a few cases where variant forms are found.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We have described two data structures for language modeling that achieve substantial reductions in time and memory cost.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The logistic function, whose outputs are in [0, 1], forces pp(s, t) <_ po(s, t).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"As you can see in the figure, the accuracy for the domain is quite high except for the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set, which contains various expressions representing different relationships for an IE application.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"6 Our knowledge sources return some sort of probability estimate, although in some cases this estimate is not especially well-principled (e.g., the Recency KS).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Given a key k, it estimates the position If the estimate is exact (A[pivot] = k), then the algorithm terminates succesfully.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"(f1; ;mg n fl1g ; l) 3 (f1; ;mg n fl; l1; l2g ; l0) !","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The content does not necessarily reflect the views of the U.S. Government, and no official endorsement should be inferred.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The larger sets are more accurate than the small sets.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"If we remove this sample from the evaluation, then the ATB type-level error rises to only 37.4% while the n-gram error rate increases to 6.24%.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The final strong hypothesis, denoted 1(x), is then the sign of a weighted sum of the weak hypotheses, 1(x) = sign (Vii atht(x)), where the weights at are determined during the run of the algorithm, as we describe below.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Consequently, we implemented our own annotation tool ConAno in Java (Stede, Heintze 2004), which provides specifically the functionality needed for our purpose.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We use a simple TF\/IDF method to measure the topicality of words.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"example, in Northern Mandarin dialects there is a morpheme -r that attaches mostly to nouns, and which is phonologically incorporated into the syllable to which it attaches: thus men2+r (door+R) 'door' is realized as mer2.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Lexical-knowledge-based approaches that include statistical information generally presume that one starts with all possible segmentations of a sentence, and picks the best segmentation from the set of possible segmentations using a probabilistic or cost\u00c2\u00ad based scoring mechanism.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"In the cases where isolated constituent precision is larger than 0.5 the affected portion of the hypotheses is negligible.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Future work should also extend the approach to build a complete named entity extractor - a method that pulls proper names from text and then classifies them.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"In line with perplexity results from Table 1, the PROBING model is the fastest followed by TRIE, and subsequently other packages.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Extract NE pair instances with contexts From the four years of newspaper corpus, we extracted 1.9 million pairs of NE instances.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"AdaBoost was first introduced in (Freund and Schapire 97); (Schapire and Singer 98) gave a generalization of AdaBoost which we will use in this paper.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"At this stage the lattice path corresponds to segments only, with no PoS assigned to them.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This is manifest in the lexical choices but 1 www.coli.unisb.de\/\u00e2\u0088\u00bcthorsten\/tnt\/ Dagmar Ziegler is up to her neck in debt.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"With a good hash function, collisions of the full 64bit hash are exceedingly rare: one in 266 billion queries for our baseline model will falsely find a key not present.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Roughly speaking, we say that a tee set contains trees with dependent paths if there are two paths p., = vim., and g., = in each 7 E r such that v., is some, possibly empty, shared initial subpath; v., and wi are not bounded in length; and there is some &quot;dependence&quot; (such as equal length) between the set of all v., and w., for each 7 Er.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The test accuracy more or less asymptotes.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Finally we show the combining techniques degrade very little when a poor parser is added to the set.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"31 75.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Our \u201cProjection\u201d baseline is able to benefit from the bilingual information and greatly improves upon the monolingual baselines, but falls short of the \u201cNo LP\u201d model by 2.5% on an average.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(2) Description Source Location Names http:\/\/www.timeanddate.com http:\/\/www.cityguide.travel-guides.com http:\/\/www.worldtravelguide.net Corporate Names http:\/\/www.fmlx.com Person First Names http:\/\/www.census.gov\/genealogy\/names Person Last Names Table 2: Sources of Dictionaries The McCann family . . .","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We have not yet tried this.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Of course, we.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"SSER: subjective sentence error rate: For a more detailed analysis, the translations are judged by a human test person.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Sorted arrays store key-value pairs in an array sorted by key, incurring no space overhead.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"A modified language model probability p\u00c3\u0086(eje0; e00) is defined as follows: p\u00c3\u0086(eje0; e00) =  1:0 if \u00c3\u0086 = 0 p(eje0; e00) if \u00c3\u0086 = 1 : We associate a distribution p(\u00c3\u0086) with the two cases \u00c3\u0086 = 0 and \u00c3\u0086 = 1 and set p(\u00c3\u0086 = 1) = 0:7.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"By definition, each existential NP uniquely specifies an object or concept, so we can infer that all instances of the same existential NP are coreferent (e.g., \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d always refers to the same entity).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The first concerns how to deal with ambiguities in segmentation.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Using structural information As was explained in the results section, we extracted examples like \u00e2\u0080\u009cSmith estimates Lotus\u00e2\u0080\u009d, from a sentence like \u00e2\u0080\u009cMr.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Our analysis begins with a description of syntactic ambiguity in unvocalized MSA text (\u00c2\u00a72).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"2.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"2.3 Rhetorical structure.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Beyond optimizing the memory size of TRIE, there are alternative data structures such as those in Guthrie and Hepple (2010).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"For the inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J), we drop the dependence on the target sentence length I. 2.2 Word Joining.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Such discrepancies can be aligned via an intermediate level of PoS tags.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"These are written to the state s(wn1) and returned so that they can be used for the following query.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Given an anaphor and candidate, BABAR checks (1) whether the semantic classes of the anaphor intersect with the semantic expectations of the caseframe that extracts the candidate, and (2) whether the semantic classes of the candidate intersect with the semantic ex pectations of the caseframe that extracts the anaphor.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Lattice parsing (Chappelier et al., 1999) is an alternative to a pipeline that prevents cascading errors by placing all segmentation options into the parse chart.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Again we notice that the isolated constituent precision is larger than 0.5 only in those partitions that contain very few samples.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"As a result, its POS tag needs to be induced in the \u201cNo LP\u201d case, while the 11A word level paired-t-test is significant at p < 0.01 for Danish, Greek, Italian, Portuguese, Spanish and Swedish, and p < 0.05 for Dutch. correct tag is available as a constraint feature in the \u201cWith LP\u201d case.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"With the same Corporate- Suffix-List and Person-Prefix-List used in local features, for a token seen elsewhere in the same document with one of these suffixes (or prefixes), another feature Other-CS (or Other-PP) is set to 1.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"A contextual role represents the role that a noun phrase plays in an event or relationship.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"3 60.7 50.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Here, \u00e2\u0080\u009cEG\u00e2\u0080\u009d represents \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The approach recursively evaluates a quantity Q(C; j), where C is the set of already visited cities and sj is the last visited city.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Figure 1 shows an example.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The Potsdam Commentary Corpus","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"D o m ai n # of ph ras es t o t a l p h r a s e s ac cu ra cy C C 7 o r m o r e 1 0 5 8 7 . 6 % 6 o r l e s s 1 0 6 6 7 . 0 % P C 7 o r m o r e 3 5 9 9 9 . 2 % 6 o r l e s s 2 5 5 6 5 . 1 % Table 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Table 4 shows translation results for the three approaches.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Our results show that BABAR achieves good performance in both domains, and that the contextual role knowledge improves performance, especially on pronouns.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"One of such approaches uses comparable documents, which are sets of documents whose content are found\/known to be almost the same, such as different newspaper stories about the same event [Shinyama and Sekine 03] or different translations of the same story [Barzilay 01].","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"As a result, Habash et al.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Evaluation results for links","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"4 Evaluation Results.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Again, the monotone search performs worst.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"37 84.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For each domain, we created a blind test set by manually annotating 40 doc uments with anaphoric chains, which represent sets of m3 (S) = ) X \u00e2\u0088\u00a9Y =S 1 \u00e2\u0088\u0092 ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) (1) noun phrases that are coreferent (as done for MUC6 (MUC6 Proceedings, 1995)).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"com t 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The belief value that would have been assigned to the intersection of these sets is .60*.70=.42, but this belief has nowhere to go because the null set is not permissible in the model.7 So this probability mass (.42) has to be redistributed.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The new algorithm, which we call CoBoost, uses labeled and unlabeled data and builds two classifiers in parallel.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The results, along with the total number of phrases, are shown in Table 1.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Tokens tagged as PUNC are not discarded unless they consist entirely of punctuation.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But Rehbein and van Genabith (2007) showed that Evalb should not be used as an indication of real difference\u00e2\u0080\u0094 or similarity\u00e2\u0080\u0094between treebanks.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"However, their system is a hybrid of hand-coded rules and machine learning methods.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"2.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"These systems rely on a training corpus that has been manually annotated with coreference links.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"m(S) represents the belief that the correct hypothesis is included in S. The model assumes that evidence also arrives as a probability density function (pdf) over sets of hypotheses.6 Integrating new evidence into the existing model is therefore simply a matter of defining a function to merge pdfs, one representing the current belief system and one representing the beliefs of the new evidence.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(a) 1 \u00c2\u00a7 . ;m t 7 leO z h e 4 pil m a 3 lu 4 sh an g4 bi ng 4 t h i s CL (assi fier) horse w ay on sic k A SP (ec t) 'This horse got sick on the way' (b) 1\u00c2\u00a7: . til y zhe4 tiao2 ma3lu4 hen3 shao3 this CL road very few 'Very few cars pass by this road' :$ chel jinglguo4 car pass by 2.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In order to evaluate and advance this approach, it helps to feed into the knowledge base data that is already enriched with some of the desired information \u00e2\u0080\u0094 as in PCC.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We have not to date explored these various options.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"IdentiFinder ' 99' s results are considerably better than IdentiFinder ' 97' s. IdentiFinder' s performance in MUC7 is published in (Miller et al., 1998).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Training under this model involves estimation of parameter values for P(y), P(m) and P(x I y).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Space- or punctuation-delimited * 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A totally non\u00c2\u00ad stochastic rule-based system such as Wang, Li, and Chang's will generally succeed in such cases, but of course runs the risk of overgeneration wherever the single-hanzi word is really intended.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"This is an iterative method that improves the estimation of the parameters at each iteration.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The bias of automatic methods in favor of statistical systems seems to be less pronounced on out-of-domain test data.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The weak hypothesis chosen was then restricted to be a predictor in favor of this label.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Note that since all English vertices were extracted from the parallel text, we will have an initial label distribution for all vertices in Ve.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"In future work we plan to try this approach with more competitive SMT systems, and to extend instance weighting to other standard SMT components such as the LM, lexical phrase weights, and lexicalized distortion.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Instead of offsetting new topics with punctuation, writers of MSA in sert connectives such as \u00ef\u00bf\u00bd wa and \u00ef\u00bf\u00bd fa to link new elements to both preceding clauses and the text as a whole.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Our analysis begins with a description of syntactic ambiguity in unvocalized MSA text (\u00c2\u00a72).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Le m\u00b4edicament de r\u00b4ef\u00b4erence de Silapo est EPREX\/ERYPO, qui contient de l\u2019\u00b4epo\u00b4etine alfa.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In our implementation, we make perhaps the simplest choice of weak hypothesis.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We checked whether the discovered links are listed in WordNet.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"To evaluate proper-name identification, we randomly se\u00c2\u00ad lected 186 sentences containing 12,000 hanzi from our test corpus and segmented the text automatically, tagging personal names; note that for names, there is always a sin\u00c2\u00ad gle unambiguous answer, unlike the more general question of which segmentation is correct.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Linear weights are difficult to incorporate into the standard MERT procedure because they are \u201chidden\u201d within a top-level probability that represents the linear combination.1 Following previous work (Foster and Kuhn, 2007), we circumvent this problem by choosing weights to optimize corpus loglikelihood, which is roughly speaking the training criterion used by the LM and TM themselves.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The work of the second author as well as collaboration visits to Israel was financed by NWO, grant number 017.001.271.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"3) A tight coupling with the speech recognizer output.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"In fact, we found that enabling IRSTLM\u2019s cache made it slightly slower, so results in Table 1 use IRSTLM without caching.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"3.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Evaluation We use 8 different measures to evaluate the performance of our system on the joint disambiguation task.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In our situation, the competing hypotheses are the possible antecedents for an anaphor.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We do not experiment with models larger than physical memory in this paper because TPT is unreleased, factors such as disk speed are hard to replicate, and in such situations we recommend switching to a more compact representation, such as RandLM.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We show the results of three of the experiments we conducted to measure isolated constituent precision under various partitioning schemes.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"4.5 Transliterations of Foreign Words.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Table 2 shows results for both settings and all methods described in sections 2 and 3.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We use label propagation in two stages to generate soft labels on all the vertices in the graph.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Each distance in the traveling salesman problem now corresponds to the negative logarithm of the product of the translation, alignment and language model probabilities.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This is motivated by taking \u03b2 po(s|t) to be the parameters of a Dirichlet prior on phrase probabilities, then maximizing posterior estimates p(s|t) given the IN corpus.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Nu mb er filters candidate if number doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Je voudrais pr\u00b4eciser, a` l\u2019adresse du commissaire Liikanen, qu\u2019il n\u2019est pas ais\u00b4e de recourir aux tribunaux nationaux.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Reading the following record\u2019s offset indicates where the block ends.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Our code is thread-safe, and integrated into the Moses, cdec, and Joshua translation systems.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Vocabulary lookup is a sorted array of 64-bit word hashes.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Variants of alif are inconsistently used in Arabic texts.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"11 taTweel (-) is an elongation character used in Arabic script to justify text.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"each word in the lexicon whether or not each string is actually an instance of the word in question.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"And then there are decisions that systems typically hard-wire, because the linguistic motivation for making them is not well understood yet.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Once again we present both a non-parametric and a parametric technique for this task.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"8 57.3 +F EA TS be st me dia n 50.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"This overview is illustrated in Figure 1.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"But diacritics are not present in unvocalized text, which is the standard form of, e.g., news media documents.3 VBD she added VP PUNC S VP VBP NP ...","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Hence we decided to select ten commentaries to form a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099, for which the entire range of annotation levels was realized, so that experiments with multi-level querying could commence.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The domains are the linguistic spans that are to receive an IS-partitioning, and the units are the (smaller) spans that can play a role as a constituent of such a partitioning.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"At each iteration the algorithm increases the number of rules, while maintaining a high level of agreement between the spelling and contextual decision lists.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Memory usage is the same as with binary search and lower than with set.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We define an ATM, M, recognizing a language generated by a grammar, G, having the properties discussed in Section 43.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The algorithm can be viewed as heuristically optimizing an objective function suggested by (Blum and Mitchell 98); empirically it is shown to be quite successful in optimizing this criterion.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(In this figure eps is c) be implemented, though, such as a maximal-grouping strategy (as suggested by one reviewer of this paper); or a pairwise-grouping strategy, whereby long sequences of unattached hanzi are grouped into two-hanzi words (which may have some prosodic motivation).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"One key issue here is to seek a discourse-based model of information structure.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"(Brandt 1996) extended these ideas toward a conception of kommunikative Gewichtung (\u00e2\u0080\u0098communicative-weight assignment\u00e2\u0080\u0099).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Evaluation results for links","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"As was explained in the results section, \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d are not desirable keywords in the CC-domain.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Its success depends on the two domains being relatively close, and on the OUT corpus not being so large as to overwhelm the contribution of IN.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Performance typically stabilizes across languages after only a few number of iterations.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"BABAR uses a named entity recognizer to identify proper names that refer to people and companies.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Formally, we define dependency graphs as follows: 3.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"(Again, the goal of also in structural features.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"For this experiment, we compare our model with the uniform tag assignment prior (1TW) with the learned prior (+PRIOR).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"What is important and is not expressed by the notation is the so-called coverage constraint: each source position j should be 'hit' exactly once by the path of the inverted alignment bI 1 = b1:::bi:::bI . Using the inverted alignments in the maximum approximation, we obtain as search criterion: max I (p(JjI) max eI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) max bI 1 I Yi=1 [p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])) = = max I (p(JjI) max eI 1;bI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])); where the two products over i have been merged into a single product over i. p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) is the trigram language model probability.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, if {N P1, N P2, N P3} are all coreferent, then each NP must be linked to one of the other two NPs.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"These 140 NE categories are designed by extending MUC\u00e2\u0080\u0099s 7 NE categories with finer sub-categories (such as Company, Institute, and Political Party for Organization; and Country, Province, and City for Location) and adding some new types of NE categories (Position Title, Product, Event, and Natural Object).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Particles are uninflected.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"There is no global pruning.","target_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"All commentaries have been annotated with rhetorical structure, using RSTTool4 and the definitions of discourse relations provided by Rhetorical Structure Theory (Mann, Thompson 1988).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Note that in our construction arcs can never cross token boundaries.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Finally, we find links between sets of phrases, based on the NE instance pair data (for example, different phrases which link \u00e2\u0080\u009cIBM\u00e2\u0080\u009d and \u00e2\u0080\u009cLotus\u00e2\u0080\u009d) (Step 4).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, kidnapping victims should be extracted from the subject of the verb \u00e2\u0080\u009ckidnapped\u00e2\u0080\u009d when it occurs in the passive voice (the shorthand representation of this pattern would be \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The baseline simply retains the original labels for all arcs, regardless of whether they have been lifted or not, and the number of distinct labels is therefore simply the number n of distinct dependency types.2 In the first encoding scheme, called Head, we use a new label d\u2191h for each lifted arc, where d is the dependency relation between the syntactic head and the dependent in the non-projective representation, and h is the dependency relation that the syntactic head has to its own head in the underlying structure.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"4 70.4 46.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"As a result, its POS tag needs to be induced in the \u201cNo LP\u201d case, while the 11A word level paired-t-test is significant at p < 0.01 for Danish, Greek, Italian, Portuguese, Spanish and Swedish, and p < 0.05 for Dutch. correct tag is available as a constraint feature in the \u201cWith LP\u201d case.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.5 Connectives with scopes.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"For example, ... fraud related to work on a federally funded sewage plant in Georgia In this case, Georgia is extracted: the NP containing it is a complement to the preposition in; the PP headed by in modifies the NP a federally funded sewage plant, whose head is the singular noun plant.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This style of naming is never required and seems to be losing currency.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Besides information structure, the second main goal is to enhance current models of rhetorical structure.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Keywords with more than one word In the evaluation, we explained that \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d are considered paraphrases.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Moses sets the cache size parameter to 50 so we did as well; the resulting cache size is 2.82 GB.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Table 1: Syntactic Seeding Heuristics BABAR\u00e2\u0080\u0099s reliable case resolution heuristics produced a substantial set of anaphor\/antecedent resolutions that will be the training data used to learn contextual role knowledge.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In various dialects of Mandarin certain phonetic rules apply at the word.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In the cotraining case, (Blum and Mitchell 98) argue that the task should be to induce functions Ii and f2 such that So Ii and 12 must (1) correctly classify the labeled examples, and (2) must agree with each other on the unlabeled examples.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Mi(c) is a binary function returning t when parser i (from among the k parsers) suggests constituent c should be in the parse.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The less favored reading may be selected in certain contexts, however; in the case of , for example, the nominal reading jiang4 will be selected if there is morphological information, such as a following plural affix ir, menD that renders the nominal reading likely, as we shall see in Section 4.3.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"For each source word f, the list of its possible translations e is sorted according to p(fje) puni(e), where puni(e) is the unigram probability of the English word e. It is su\u00c3\u0086cient to consider only the best 50 words.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In this subsection, we will report the results of the experiment, in terms of the number of words, phrases or clusters.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"It can be shown that this objective is convex in q.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The 14 general-language features embody straightforward cues: frequency, \u201ccentrality\u201d as reflected in model scores, and lack of burstiness.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We have argued that the proposed method performs well.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"SRILM\u2019s compact variant, IRSTLM, MITLM, and BerkeleyLM\u2019s sorted variant are all based on this technique.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(1991}, Gu and Mao (1994), and Nie, Jin, and Hannan (1994).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"There are 13 types of extensions needed to describe the verbgroup reordering.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The average agreement among the human judges is .76, and the average agreement between ST and the humans is .75, or about 99% of the interhuman agreement.15 One can better visualize the precision-recall similarity matrix by producing from that matrix a distance matrix, computing a classical metric multidimensional scaling (Torgerson 1958; Becker, Chambers, Wilks 1988) on that dis\u00c2\u00ad tance matrix, and plotting the first two most significant dimensions.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Unigram lookup is dense so we use an array of probability and backoff values.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Search t0 CPU time #search mWER Method [sec] error [%] QmS 0.0 0.07 108 42:6 1.0 0.13 85 37:8 2.5 0.35 44 36:6 5.0 1.92 4 34:6 10.0 10.6 0 34:5 IbmS 0.0 0.14 108 43:4 1.0 0.3 84 39:5 2.5 0.8 45 39:1 5.0 4.99 7 38:3 10.0 28.52 0 38:2 Table 6 shows example translations obtained by the three different approaches.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"For one thing, it is not clear who is to receive settlements or what should happen in case not enough teachers accept the offer of early retirement.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Sheffield Guthrie and Hepple (2010) explore several randomized compression techniques, but did not release code.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"In contrast, NNP (proper nouns) form a large portion of vocabulary.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"For even larger models, we recommend RandLM; the memory consumption of the cache is not expected to grow with model size, and it has been reported to scale well.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Altun et al. (2005) proposed a technique that uses graph based similarity between labeled and unlabeled parts of structured data in a discriminative framework for semi-supervised learning.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"It was also proposed to allow annotators to skip sentences that they are unable to judge.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In our work, we demonstrate that using a simple na\u00c2\u00a8\u00c4\u00b1veBayes approach also yields substantial performance gains, without the associated training complexity.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Altun et al. (2005) proposed a technique that uses graph based similarity between labeled and unlabeled parts of structured data in a discriminative framework for semi-supervised learning.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"1).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Further, we report current resident memory and peak virtual memory because these are the most applicable statistics provided by the kernel.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We include a list of per-category results for selected phrasal labels, POS tags, and dependencies in Table 8.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This differs from other implementations (Stolcke, 2002; Pauls and Klein, 2011) that use hash tables as nodes in a trie, as explained in the next section.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Morphological processes in Semitic languages deliver space-delimited words which introduce multiple, distinct, syntactic units into the structure of the input sentence.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"3) A tight coupling with the speech recognizer output.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Unfortunately, we were unable to correctly run the IRSTLM quantized variant.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(3)) to be defined over unlabeled as well as labeled instances.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The general idea for the knowledge- based part is to have the system use as much information as it can find at its disposal to produce a target representation as specific as possible and as underspecified as necessary.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Corporate Suffixes and Person Prefixes of Other Occurrences (CSPP): If McCann has been seen as Mr. McCann somewhere else in the document, then one would like to give person a higher probability than organization.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Both BBN and NYU have tagged their own data to supplement the official training data.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(Yarowsky 95) describes an algorithm for word-sense disambiguation that exploits redundancy in contextual features, and gives impressive performance.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Other errors include NE tagging errors and errors due to a phrase which includes other NEs.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"BABAR achieved recall in the 4250% range for both domains, with 76% precision overall for terrorism and 87% precision for natural disasters.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Our final average POS tagging accuracy of 83.4% compares very favorably to the average accuracy of Berg-Kirkpatrick et al.\u2019s monolingual unsupervised state-of-the-art model (73.0%), and considerably bridges the gap to fully supervised POS tagging performance (96.6%).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Thus at each iteration the algorithm is forced to pick features for the location, person and organization in turn for the classifier being trained.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The NE tagger is a rule-based system with 140 NE categories [Sekine et al. 2004].","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For definite NPs, the results are a mixed bag: some knowledge sources increased recall a little, but at the expense of some precision.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"97 78.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We proposed an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The zone to which a token belongs is used as a feature.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Examples will usually be accompanied by a translation, plus a morpheme-by-morpheme gloss given in parentheses whenever the translation does not adequately serve this purpose.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"(2010) consistently outperforms ours on English, we obtain substantial gains across other languages.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We were intentionally lenient with our baselines: bilingual information by projecting POS tags directly across alignments in the parallel data.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The commentaries in PCC are all of roughly the same length, ranging from 8 to 10 sentences.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Our results show that MENERGI performs as well as IdentiFinder when trained on comparable amount of training data.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"2.2.3 Lexical Caseframe Expectations The second type of contextual role knowledge learned by BABAR is Lexical Caseframe Expectations, which are used by the CFLex knowledge source.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Because these two words have identical complements, syntax rules are typically unhelpful for distinguishing between them.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Time includes all queries but excludes random number generation and data structure population.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"However their work did not consider other types of lexical expectations (e.g., PP arguments), semantic expectations, or context comparisons like our case- frame network.(Niyu et al., 1998) used unsupervised learning to ac quire gender, number, and animacy information from resolutions produced by a statistical pronoun resolver.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We would also like to thank Amarnag Subramanya for helping us with the implementation of label propagation and Shankar Kumar for access to the parallel data.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"1 2 3.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"From the point of view of computational implementation this can be problematic, since the inclusion of non-projective structures makes the parsing problem more complex and therefore compromises efficiency and in practice also accuracy and robustness.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In the ATB, :: b asta\u00e2\u0080\u0099adah is tagged 48 times as a noun and 9 times as verbal noun.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"5.2 Discussion.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"7.96 5.55 1 l...................................................................................................................................................................................................J..","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Because of this threshold, very few NE instance pairs could be used and hence the variety of phrases was also limited.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"However, if we consider precision, recall and Fmeasure on non-projective dependencies only, as shown in Table 6, some differences begin to emerge.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Alon Lavie advised on this work.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"When aligning the words in parallel texts (for language pairs like SpanishEnglish, French-English, ItalianGerman,...), we typically observe a strong localization effect.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"For graph propagation, the hyperparameter v was set to 2 x 10\u22126 and was not tuned.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses the log-likelihood statistic (Dunning, 1993) to evaluate the strength of a co-occurrence relationship.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Corporate Suffixes and Person Prefixes of Other Occurrences (CSPP): If McCann has been seen as Mr. McCann somewhere else in the document, then one would like to give person a higher probability than organization.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2010) and the posterior regular- ization HMM of Grac\u00c2\u00b8a et al.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Finally, we thank Kuzman Ganchev and the three anonymous reviewers for helpful suggestions and comments on earlier drafts of this paper.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Consider first the examples in (2).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"evaluated to account for the same fraction of the data.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"27 80.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We concentrate on those sets.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Step 4.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For MUC6, for example, Corporate- Suffix-List is made up of ltd., associates, inc., co, corp, ltd, inc, committee, institute, commission, university, plc, airlines, co., corp. and Person-Prefix- List is made up of succeeding, mr., rep., mrs., secretary, sen., says, minister, dr., chairman, ms. . For a token that is in a consecutive sequence of init then a feature Corporate-Suffix is set to 1.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"This &quot;default&quot; feature type has 100% coverage (it is seen on every example) but a low, baseline precision.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Finally, the concatenated 5 * 20% output is used to train the reference resolution component.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Although these authors report better gains than ours, they are with respect to a non-adapted baseline.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Table 4 shows how much the Bayes switching technique uses each of the parsers on the test set.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"From the discussion so far it is clear that a number of formalisms involve some type of context-free rewriting (they have derivation trees that are local sets).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Person\u00e2\u0080\u009d domain, 618 keywords are found, and in the \u00e2\u0080\u009cCountry \u00e2\u0080\u0093 Country\u00e2\u0080\u009d domain, 303 keywords are found.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Finally, we would like to note that it is possible to devise similar algorithms based with other objective functions than the one given in Equ.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"RandLM 0.2 (Talbot and Osborne, 2007) stores large-scale models in less memory using randomized data structures.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"IRST is not threadsafe.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Then, it can be verified that We can now derive the CoBoost algorithm as a means of minimizing Zco.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Again, famous place names will most likely be found in the dictionary, but less well-known names, such as 1PM\u00c2\u00b1 R; bu4lang3-shi4wei2-ke4 'Brunswick' (as in the New Jersey town name 'New Brunswick') will not generally be found.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.2 Syntactic structure.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"needs to be in initCaps to be considered for this feature.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The PROBING model is 2.4 times as fast as the fastest alternative, SRILM, and uses less memory too.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"We often encounter sentences that are highly ambiguous in themselves, without some prior knowledge of the entities concerned.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"5.2 Setup.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Equation 2 is an estimate of the conditional probability of the label given the feature, P(yjx).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(3) shows learning curves for CoBoost.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"suffixes (e.g., \u00ef\u00bf\u00bd=) Other notable parameters are second order vertical Markovization and marking of unary rules.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Each distance in the traveling salesman problem now corresponds to the negative logarithm of the product of the translation, alignment and language model probabilities.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We refer to this process as Reliable Case Resolution because it involves finding cases of anaphora that can be easily resolved with their antecedents.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We use v1.0 mainly because previous studies on joint inference reported results w.r.t. v1.0 only.5 We expect that using the same setup on v2.0 will allow a crosstreebank comparison.6 We used the first 500 sentences as our dev set and the rest 4500 for training and report our main results on this split.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Other strategies could readily 6 As a reviewer has pointed out, it should be made clear that the function for computing the best path is. an instance of the Viterbi algorithm.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"However, when we pre- tag the input\u00e2\u0080\u0094as is recommended for English\u00e2\u0080\u0094 we notice a 0.57% F1 improvement.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A Brief Introduction to the Chinese Writing System Most readers will undoubtedly be at least somewhat familiar with the nature of the Chinese writing system, but there are enough common misunderstandings that it is as well to spend a few paragraphs on properties of the Chinese script that will be relevant to topics discussed in this paper.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Each word is terminated by an arc that represents the transduction between f and the part of speech of that word, weighted with an estimated cost for that word.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Fortunately, we were able to obtain a copy of the full set of sentences from Chang et al. on which Wang, Li, and Chang tested their system, along with the output of their system.18 In what follows we will discuss all cases from this set where our performance on names differs from that of Wang, Li, and Chang.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"This is the form of recursive levels in iDafa constructs.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We claim that no particular morphological segmentation is a-priory more likely for surface forms before exploring the compositional nature of syntactic structures, including manifestations of various long-distance dependencies.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"A Stochastic Finite-State Word-Segmentation Algorithm for Chinese","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We consider the unsupervised POS induction problem without the use of a tagging dictionary.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":".","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The only supervision is in the form of 7 seed rules (namely, that New York, California and U.S. are locations; that any name containing Mr is a person; that any name containing Incorporated is an organization; and that I.B.M. and Microsoft are organizations).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The similar explanation applies to the link to the \u00e2\u0080\u009cstake\u00e2\u0080\u009d set.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Here, all tokeninternal collocations of tags unseen in our training data are pruned away.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Since these are distinct syntactic units, they are typically segmented.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Furthermore, by inverting the transducer so that it maps from phonemic transcriptions to hanzi sequences, one can apply the segmenter to other problems, such as speech recognition (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Our original hope in combining these parsers is that their errors are independently distributed.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"All of these systems were run on data that was not seen during their development.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Figure 2 shows timing results.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"0 D\/ nc 5.0 The minimal dictionary encoding this information is represented by the WFST in Figure 2(a).","label":0},{"source_text":"There is no global pruning.","target_text":"The negative logarithm of t0 is reported.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Table 2 shows these similarity measures.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"As seen by the drop in average individual parser performance baseline, the introduced parser does not perform very well.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"With the additional assumption that (s, t) can be restricted to the support of co(s, t), this is equivalent to a \u201cflat\u201d alternative to (6) in which each non-zero co(s, t) is set to one.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The following three sections elaborate these different stages is more detail.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"For example, the two NEs \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d and \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d have the following contexts.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Our results show that MENERGI performs as well as IdentiFinder when trained on comparable amount of training data.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"7 Unlike Dickinson (2005), we strip traces and only con-.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"None of the models we have presented utilize features associated with a particular constituent (i.e. the label, span, parent label, etc.) to influence parser preference.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"There is a fairly large body of work on SMT adaptation.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"As discussed in more detail in \u00a73, we use two types of vertices in our graph: on the foreign language side vertices correspond to trigram types, while the vertices on the English side are individual word types.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":".","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"rently, some annotations (in particular the connectives and scopes) have already moved beyond the core corpus; the others will grow step by step.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"One conclusion drawn from this annotation effort was that for humans and machines alike, 2 www.sfs.nphil.unituebingen.de\/Elwis\/stts\/ stts.html 3 www.coli.unisb.de\/sfb378\/negra-corpus\/annotate.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Hence we decided to select ten commentaries to form a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099, for which the entire range of annotation levels was realized, so that experiments with multi-level querying could commence.","label":1},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This section describes an algorithm based on boosting algorithms, which were previously developed for supervised machine learning problems.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For MUC7, there are also no published results on systems trained on only the official training data of 200 aviation disaster articles.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We again adopt an approach where we alternate between two classifiers: one classifier is modified while the other remains fixed.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In many cases, there is an even stronger restriction: over large portions of the source string, the alignment is monotone.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For example, it is well-known that one can build a finite-state bigram (word) model by simply assigning a state Si to each word Wi in the vocabulary, and having (word) arcs leaving that state weighted such that for each Wj and corresponding arc aj leaving Si, the cost on aj is the bigram cost of WiWj- (Costs for unseen bigrams in such a scheme would typically be modeled with a special backoff state.)","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"A morphological analyzer M : W\u2014* L is a function mapping sentences in Hebrew (W E W) to their corresponding lattices (M(W) = L E L).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The accuracy of the sets in representing paraphrase ranged from 73% to 99%, depending on the NE categories and set sizes; the accuracy of the links for two evaluated domains was 73% and 86%.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The method uses a &quot;soft&quot; measure of the agreement between two classifiers as an objective function; we described an algorithm which directly optimizes this function.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"A number of PCC commentaries will be read by professional news speakers and prosodic features be annotated, so that the various annotation layers can be set into correspondence with intonation patterns.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We report results for the best and median hyperparameter settings obtained in this way.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The new algorithm, which we call CoBoost, uses labeled and unlabeled data and builds two classifiers in parallel.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Statistical NERs usually find the sequence of tags that maximizes the probability , where is the sequence of words in a sentence, and is the sequence of named-entity tags assigned to the words in . Attempts have been made to use global information (e.g., the same named entity occurring in different sentences of the same document), but they usually consist of incorporating an additional classifier, which tries to correct the errors in the output of a first NER (Mikheev et al., 1998; Borthwick, 1999).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"In Table 7 we give results for several evaluation metrics.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Whereas Arabic linguistic theory as Saddam (a) Reference (b) Stanford signs (1) and (2) to the class of pseudo verbs 01 +i J>1\u00ef\u00bf\u00bd inna and her sisters since they can beinflected, the ATB conventions treat (2) as a com plementizer, which means that it must be the head of SBAR.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"This paper does not necessarily reflect the position of the U.S. Government.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The secondary classifier in (Borthwick, 1999) uses information not just from the current article, but also from the whole test corpus, with an additional feature that indicates if the information comes from the same document or from another document.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"From the point of view of computational implementation this can be problematic, since the inclusion of non-projective structures makes the parsing problem more complex and therefore compromises efficiency and in practice also accuracy and robustness.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.1.1 Lexical Seeding It is generally not safe to assume that multiple occurrences of a noun phrase refer to the same entity.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"However, the learning curves in Figure 3 show that the Berkeley parser does not exceed our manual grammar by as wide a margin as has been shown for other languages (Petrov, 2009).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"While Gan's system incorporates fairly sophisticated models of various linguistic information, it has the drawback that it has only been tested with a very small lexicon (a few hundred words) and on a very small test set (thirty sentences); there is therefore serious concern as to whether the methods that he discusses are scalable.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Our full model (\u201cWith LP\u201d) outperforms the unsupervised baselines and the \u201cNo LP\u201d setting for all languages.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(1992).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Section 2.2 then describes our representation for contextual roles and four types of contextual role knowledge that are learned from the training examples.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"They cluster NE instance pairs based on the words in the contexts using a bag- of-words method.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"One hopes that such a corpus will be forth\u00c2\u00ad coming.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"9 65.5 46.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"95 76.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"kann 7.nicht 8.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"From the definition of TAG's, it follows that the choice of adjunction is not dependent on the history of the derivation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"As we have seen, the lexicon of basic words and stems is represented as a WFST; most arcs in this WFST represent mappings between hanzi and pronunciations, and are costless.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We will show that by giving the first model some global features, MENERGI outperforms Borthwick' s reference resolution classifier.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"At first glance, this seems only peripherally related to our work, since the specific\/general distinction is made for features rather than instances.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Log-linear combination (loglin) improves on this in all cases, and also beats the pure IN system.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"With this restriction the resulting tree sets will have independent paths.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"4.2 A Sample Segmentation Using Only Dictionary Words Figure 4 shows two possible paths from the lattice of possible analyses of the input sentence B X:\u00c2\u00a5 .:.S:P:l 'How do you say octopus in Japanese?' previously shown in Figure 1.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"2.1 Inverted Alignments.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"We propose maximizing , where is the sequence of named- entity tags assigned to the words in the sentence , and is the information that can be extracted from the whole document containing . Our system is built on a maximum entropy classifier.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Using this encoding scheme, the arc from je to Z in Figure 2 would be assigned the label AuxP\u2191Sb (signifying an AuxP that has been lifted from a Sb).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Our model outperforms theirs on four out of five languages on the best hyperparameter setting and three out of five on the median setting, yielding an average absolute difference across languages of 12.9% and 3.9% for best and median settings respectively compared to their best EM or LBFGS performance.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The addition of vertical markovization enables non-pruned models to outperform all previously reported re12Cohen and Smith (2007) make use of a parameter (\u03b1) which is tuned separately for each of the tasks.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For the PROBING implementation, hash table sizes are in the millions, so the most relevant values are on the right size of the graph, where linear probing wins.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We combined evidence from four contextual role knowledge sources with evidence from seven general knowledge sources using a DempsterShafer probabilistic model.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Different annotations of the same text are mapped into the same data structure, so that search queries can be formulated across annotation levels.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The relevance of the distinction between, say, phonological words and, say, dictionary words is shown by an example like rpftl_A :;!:Hfllil zhong1hua2 ren2min2 gong4he2-guo2 (China people republic) 'People's Republic of China.'","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Model Overview The model starts by generating a tag assignment T for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We are currently exploring such algorithms.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"It is unique, agrees with the maximum-likelihood distribution, and has the exponential form (Della Pietra et al., 1997): where refers to the outcome, the history (or context), and is a normalization function.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The P (W |T , \u00cf\u0088) term in the lexicon component now decomposes as: n P (W |T , \u00cf\u0088) = n P (Wi|Ti, \u00cf\u0088) i=1 n \u00ef\u00a3\u00ab \u00ef\u00a3\u00b6 tions are not modeled by the standard HMM, which = n \u00ef\u00a3\u00ad n P (v|\u00cf\u0088Ti f )\u00ef\u00a3\u00b8 instead can model token-level frequency.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We train linear mixture models for conditional phrase pair probabilities over IN and OUT so as to maximize the likelihood of an empirical joint phrase-pair distribution extracted from a development set.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Gather phrases using keywords Now, the keyword with the top TF\/ITF score is selected for each phrase.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"A defining characteristic of MSA is the prevalence of discourse markers to connect and subordinate words and phrases (Ryding, 2005).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The overall performance of the LTG system was outstanding, but the system consists of a sequence of many hand-coded rules and machine-learning modules.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This enables us to build a high performance NER without using separate classifiers to take care of global consistency or complex formulation on smoothing and backoff models (Bikel et al., 1997).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"This work was funded by NSF grant IRI-9502312.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Recent research has demonstrated that incorporating this sparsity constraint improves tagging accuracy.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Morphologically derived words such as, xue2shengl+men0.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"All commentaries have been annotated with rhetorical structure, using RSTTool4 and the definitions of discourse relations provided by Rhetorical Structure Theory (Mann, Thompson 1988).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"A corpus of German newspaper commentaries has been assembled and annotated with different information (and currently, to different degrees): part-of-speech, syntax, rhetorical structure, connectives, co-reference, and information structure.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We then gather all phrases with the same keyword.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Next, we describe four contextual role knowledge sources that are created from the training examples and the caseframes.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"immediately by a Romanization into the pinyin transliteration scheme; numerals following each pinyin syllable represent tones.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Unsupervised Models for Named Entity Classification Collins","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"To see this, note thai the first two terms in the above equation correspond to the function that AdaBoost attempts to minimize in the standard supervised setting (Equ.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"30 16.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Evaluation within a set The evaluation of paraphrases within a set of phrases which share a keyword is illustrated in Figure 4.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"First, any grammar must involve a finite number of elementary structures, composed using a finite number of composition operations.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Time for Moses itself to load, including loading the language model and phrase table, is included.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"[Hasegawa et al. 04] reported only on relation discovery, but one could easily acquire para phrases from the results.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Riloff and Shepherd 97) describe a bootstrapping approach for acquiring nouns in particular categories (such as &quot;vehicle&quot; or &quot;weapon&quot; categories).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We empirically control for the effect of our heuristics to make sure our pruning does not undermine the objectives of our joint task.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"It differs from previous machine learning-based NERs in that it uses information from the whole document to classify each word, with just one classifier.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, for our purposes it is not sufficient to repre\u00c2\u00ad sent the morphological decomposition of, say, plural nouns: we also need an estimate of the cost of the resulting word.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"52 77.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"It has been shown in (Chandra et al., 1981) that if M works in space log n then there is a deterministic TM which accepts the same language in polynomial time.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(Blum and Mitchell 98) go on to give PAC results for learning in the cotraining case.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"5 64.7 42.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We use w erations of sampling (see Figure 2 for a depiction).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We have not to date explored these various options.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Links can be of two different kinds: anaphoric or bridging (definite noun phrases picking up an antecedent via world-knowledge).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Set the decision list to include all rules whose (smoothed) strength is above some threshold Pmin.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Because these two words have identical complements, syntax rules are typically unhelpful for distinguishing between them.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"When a token fmnh is to be interpreted as the lexeme sequence f\/REL mnh\/VB, the analysis introduces two distinct entities, the relativizer f (\u201cthat\u201d) and the verb mnh (\u201ccounted\u201d), and not as the complex entity \u201cthat counted\u201d.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We use a squared loss to penalize neighboring vertices that have different label distributions: kqi \u2212 qjk2 = Ey(qi(y) \u2212 qj(y))2, and additionally regularize the label distributions towards the uniform distribution U over all possible labels Y.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"When OUT is large and distinct, its contribution can be controlled by training separate IN and OUT models, and weighting their combination.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The first probability is estimated from a name count in a text database, and the rest of the probabilities are estimated from a large list of personal names.n Note that in Chang et al.'s model the p(rule 9) is estimated as the product of the probability of finding G 1 in the first position of a two-hanzi given name and the probability of finding G2 in the second position of a two-hanzi given name, and we use essentially the same estimate here, with some modifications as described later on.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"9 66.4 47.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Our system fails in (a) because of$ shenl, a rare family name; the system identifies it as a family name, whereas it should be analyzed as part of the given name.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For the translation model Pr(fJ 1 jeI 1), we go on the assumption that each source word is aligned to exactly one target word.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We used C = 1.0 as the L2 regularization constant in (Eq.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"0 X u} \"' o; .2 X X><X X XX X X X X X X x X X X X X x X V X X X X .;t'*- XXX:OX X X X X X X 9 x X X XX XX X X X X X X X XXX:< X X>O<XX>!KXX XI<>< \u00c2\u00bbC X X XX :X: X X \"' X X XX >OO<X>D<XIK X X X X X X --XX\u00c2\u00bb: XXX X X\u00c2\u00bbC X X\u00c2\u00abX...C:XXX X Xll< X X ><XX>IIC:liiC:oiiiiCI--8!X:liiOC!I!S8K X X X 10 100 1000 10000 log(F)_base: R\"2=0.20 (p < 0.005) X 100000 Figure 6 Plot of log frequency of base noun, against log frequency of plural nouns.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Note that in our model, conditioned on T , there is precisely one t which has nonzero probability for the token component, since for each word, exactly one \u00ce\u00b8t has support.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Indeed, as we shall show in Section 5, even human judges differ when presented with the task of segmenting a text into words, so a definition of the criteria used to determine that a given segmentation is correct is crucial before one can interpret such measures.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"This &quot;default&quot; feature type has 100% coverage (it is seen on every example) but a low, baseline precision.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"For each language, we took the same number of sentences from the bitext as there are in its treebank, and trained a supervised feature-HMM.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"A cell in the bottom row of the parse chart is required for each potential whitespace boundary.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Two general approaches are presented and two combination techniques are described for each approach.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We use the HSPELL9 (Har\u2019el and Kenigsberg, 2004) wordlist as a lexeme-based lexicon for pruning segmentations involving invalid segments.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We represent all morphological analyses of a given utterance using a lattice structure.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":":zhong1 : 0.0 tjl :huo2 :0.0 (R:spub:\/ic of Ch:ina) + .,_,...I : jlong4 :0.0 (mUifaty genG181) 0 \u00c2\u00a3: _NC: 40.0 Figure 3 Partial Chinese Lexicon (NC = noun; NP = proper noun).c=- - I \u00e2\u0080\u00a2=- :il: .;ss:;zhangt \u00e2\u0080\u00a2 '-:.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Our representation of contextual roles is based on information extraction patterns that are converted into simple caseframes.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The weak learner for two-class problems computes a weak hypothesis h from the input space into the reals (h : 2x -4 R), where the sign4 of h(x) is interpreted as the predicted label and the magnitude I h(x)I is the confidence in the prediction: large numbers for I h(x)I indicate high confidence in the prediction, and numbers close to zero indicate low confidence.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Denote by g3(x) = Et crithl(x) , j E {1,2} the unthresholded strong-hypothesis (i.e., f3 (x) = sign(gi (x))).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For irt the Good-Turing estimate just discussed gives us an estimate of p(unseen(f,) I f,)-the probability of observing a previously unseen instance of a construction in ft given that we know that we have a construction in f,.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We will show that by giving the first model some global features, MENERGI outperforms Borthwick' s reference resolution classifier.","label":0},{"source_text":"There is no global pruning.","target_text":"In this paper, we describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"However, we needed to restrict ourselves to these languages in order to be able to evaluate the performance of our approach.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We asked six native speakers-three from Taiwan (TlT3), and three from the Mainland (M1M3)-to segment the corpus.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"31 75.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We therefore used the arithmetic mean of each interjudge precision-recall pair as a single measure of interjudge similarity.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Beyond optimizing the memory size of TRIE, there are alternative data structures such as those in Guthrie and Hepple (2010).","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Although the best published results for the Collins parser is 80% UAS (Collins, 1999), this parser reaches 82% when trained on the entire training data set, and an adapted version of Charniak\u2019s parser (Charniak, 2000) performs at 84% (Jan Haji\u02c7c, pers. comm.).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The verbal reading arises when the maSdar has an NP argument which, in vocalized text, is marked in the accusative case.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"(1998) have also used a maximum entropy classifier that uses already tagged entities to help tag other entities.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Other strategies could readily 6 As a reviewer has pointed out, it should be made clear that the function for computing the best path is. an instance of the Viterbi algorithm.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Contextual role knowledge provides evidence as to whether a candidate is a plausible antecedent for an anaphor.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Particular relations are also consistent with particular hypotheses about the segmentation of a given sentence, and the scores for particular relations can be incremented or decremented depending upon whether the segmentations with which they are consistent are \"popular\" or not.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"For a given partial hypothesis (C; j), the order in which the cities in C have been visited can be ignored (except j), only the score for the best path reaching j has to be stored.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"As one can see, only the trigrams [suo incarceramento ,], [suo iter ,] and [suo carattere ,] are connected to English words.","label":0},{"source_text":"Here we present two algorithms.","target_text":"This allow the learners to &quot;bootstrap&quot; each other by filling the labels of the instances on which the other side has abstained so far.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"e0; e are the last two target words, C is a coverage set for the already covered source positions and j is the last position visited.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"For this experiment, we compare our model with the uniform tag assignment prior (1TW) with the learned prior (+PRIOR).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"= p(fj je) max \u00c3\u0086;e00 j02Cnfjg np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j 0 )o: The DP equation is evaluated recursively for each hypothesis (e0; e; C; j).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The constituent voting and na\u00efve Bayes techniques are equivalent because the parameters learned in the training set did not sufficiently discriminate between the three parsers.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"A Brief Introduction to the Chinese Writing System Most readers will undoubtedly be at least somewhat familiar with the nature of the Chinese writing system, but there are enough common misunderstandings that it is as well to spend a few paragraphs on properties of the Chinese script that will be relevant to topics discussed in this paper.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The second experiment is limited to data from PDT.5 The training part of the treebank was projectivized under different encoding schemes and used to train memory-based dependency parsers, which were run on the test part of the treebank, consisting of 7,507 sentences and 125,713 tokens.6 The inverse transformation was applied to the output of the parsers and the result compared to the gold standard test set.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This would result in better rest cost estimation and better pruning.10 In general, tighter, but well factored, integration between the decoder and language model should produce a significant speed improvement.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The 4th block contains instance-weighting models trained on all features, used within a MAP TM combination, and with a linear LM mixture.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Matsoukas et al (2009) generalize it by learning weights on sentence pairs that are used when estimating relative-frequency phrase-pair probabilities.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Unigrams also have 64-bit overhead for vocabulary lookup.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, co-occurring caseframes may reflect synonymy (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> abducted\u00e2\u0080\u009d) or related events (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> released\u00e2\u0080\u009d).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Finally, we show that in application settings, the absence of gold segmentation lowers parsing performance by 2\u00e2\u0080\u00935% F1.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"An ATM has two types of states, existential and universal.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"This &quot;default&quot; feature type has 100% coverage (it is seen on every example) but a low, baseline precision.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"36.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We believe it is natural for authors to use abbreviations in subsequent mentions of a named entity (i.e., first \u00e2\u0080\u009cPresident George Bush\u00e2\u0080\u009d then \u00e2\u0080\u009cBush\u00e2\u0080\u009d).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"summit Sharm (a) Al-Sheikh summit Sharm (b) DTNNP Al-Sheikh in a corpus position without a bracketing label, then we also add \u00e2\u0088\u0097n, NIL) to M. We call the set of unique n-grams with multiple labels in M the variation nuclei of C. Bracketing variation can result from either annotation errors or linguistic ambiguity.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We again adopt an approach where we alternate between two classifiers: one classifier is modified while the other remains fixed.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The original tag set for the CoNLL-X Dutch data set consists of compounded tags that are used to tag multi-word units (MWUs) resulting in a tag set of over 300 tags.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"4 70.4 46.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(c) Coordination ambiguity is shown in dependency scores by e.g., \u00e2\u0088\u0097SSS R) and \u00e2\u0088\u0097NP NP NP R).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"85 82.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Automatic Paraphrase Discovery based on Context and Keywords between NE Pairs","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thatcher (1973) describes a tee pumping lemma for recognizable sets related to the string pumping lemma for regular sets.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"As is standard, we report the greedy one-to-one (Haghighi and Klein, 2006) and the many-to-one token-level accuracy obtained from mapping model states to gold POS tags.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Figure 2: Order in which source positions are visited for the example given in Fig.1.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"While possible to utilize the feature-based log-linear approach described in Berg-Kirkpatrick et al.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"10) and trained both EM and L-BFGS for 1000 iterations.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The translation of one position in the source sentence may be postponed for up to L = 3 source positions, and the translation of up to two source positions may be anticipated for at most R = 10 source positions.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The morphological anal\u00c2\u00adysis itself can be handled using well-known techniques from finite-state morphol 9 The initial estimates are derived from the frequencies in the corpus of the strings of hanzi making up.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Particles are uninflected.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"An IG can be viewed as a CFG in which each nonterminal is associated with a stack.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Evaluation We use 8 different measures to evaluate the performance of our system on the joint disambiguation task.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"both MENE and IdentiFinder used more training data than we did (we used only the official MUC 6 and MUC7 training data).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In the terrorism domain, 1600 texts were used for training and the 40 test docu X \u00e2\u0088\u00a9Y =\u00e2\u0088\u0085 All sets of hypotheses (and their corresponding belief values) in the current model are crossed with the sets of hypotheses (and belief values) provided by the new evidence.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"with the number of exactly matching guess trees.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"As shown in Table 3, the proportion of sentences containing some non-projective dependency ranges from about 15% in DDT to almost 25% in PDT.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This WFST represents the segmentation of the text into the words AB and CD, word boundaries being marked by arcs mapping between f and part-of-speech labels.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We feel that information from a whole corpus might turn out to be noisy if the documents in the corpus are not of the same genre.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The reason why we did not train with both MUC6 and MUC7 training data at the same time is because the task specifications for the two tasks are not identical.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"But for eign learners are often surprised by the verbless predications that are frequently used in Arabic.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The Penn Arabic Treebank (ATB) syntactic guidelines (Maamouri et al., 2004) were purposefully borrowed without major modification from English (Marcus et al., 1993).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Table 2 compares the performance of our system on the setup of Cohen and Smith (2007) to the best results reported by them for the same tasks.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Further, we report current resident memory and peak virtual memory because these are the most applicable statistics provided by the kernel.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Cohen and Smith (2007) later on based a system for joint inference on factored, independent, morphological and syntactic components of which scores are combined to cater for the joint inference task.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The sentence length probability p(JjI) is omitted without any loss in performance.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Future work should also extend the approach to build a complete named entity extractor - a method that pulls proper names from text and then classifies them.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We suggest that in unlexicalized PCFGs the syntactic context may be explicitly modeled in the derivation probabilities.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Two aspects of the corpus have been presented in previous papers ((Re- itter, Stede 2003) on underspecified rhetorical structure; (Stede 2003) on the perspective of knowledge-based summarization).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Note that it is in precision that our over\u00c2\u00ad all performance would appear to be poorer than the reported performance of Chang et al., yet based on their published examples, our system appears to be doing better precisionwise.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Case and Zone: If the token starts with a capital letter (initCaps), then an additional feature (init- Caps, zone) is set to 1.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Although we have not done any experiments on other languages, this way of using global features from a whole document should be applicable to other languages.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Figure 4 Input lattice (top) and two segmentations (bottom) of the sentence 'How do you say octopus in Japanese?'.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For derived words that occur in our corpus we can estimate these costs as we would the costs for an underived dictionary entry.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A consequence of the ability to generate tree sets with this property is that CC's under this definition can generate the following language which can not be generated by either TAG's or HG's.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"A final alternate approach would be to combine weighted joint frequencies rather than conditional estimates, ie: cI(s, t) + w,\\(s, t)co(, s, t), suitably normalized.5 Such an approach could be simulated by a MAP-style combination in which separate 0(t) values were maintained for each t. This would make the model more powerful, but at the cost of having to learn to downweight OUT separately for each t, which we suspect would require more training data for reliable performance.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"It is chosen such that the decisions it made in including or excluding constituents are most probable under the models for all of the parsers.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Figure 4 shows some such phrase sets based on keywords in the CC-domain.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"phrase (markContainsVerb).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Since our goal is to perform well under these measures we will similarly treat constituents as the minimal substructures for combination.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Naseem et al. (2009) and Snyder et al.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Section 4.1 explained that state s is stored by applications with partial hypotheses to determine when they can be recombined.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Removing the leaves from the resulting tree yields a parse for L under G, with the desired probabilities.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Since the lattice L for a given sentence W is determined by the morphological analyzer M we have which is precisely the formula corresponding to the so-called lattice parsing familiar from speech recognition.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"On each language we investigate the contribution of each component of our model.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Values in the trie are minimally sized at the bit level, improving memory consumption over trie implementations in SRILM, IRSTLM, and BerkeleyLM.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Ignoring the identity of the target language words e and e0, the possible partial hypothesis extensions due to the IBM restrictions are shown in Table 2.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Hieu Hoang named the code \u201cKenLM\u201d and assisted with Moses along with Barry Haddow.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"(S; C; j); Not only the coverage set C and the positions j; j0, but also the verbgroup states S; S0 are taken into account.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This overview is illustrated in Figure 1.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"60 ( P e tr o v, 2 0 0 9 ) all B e r k e l e y ( S e p . 0 9 ) B a s e l i n e 7 0 a l l G o l d P O S 70 \u00e2\u0080\u0094 \u00e2\u0080\u0094 \u00e2\u0080\u0094 0 . 8 0 9 0.839 335 0 . 7 9","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"It is a relatively frequent word in the domain, but it can be used in different extraction scenarios.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"If the parse contains productions from outside our grammar the machine has no direct method for handling them (e.g. the resulting database query may be syntactically malformed).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Entries for 2 < n < N store a vocabulary identifier, probability, backoff, and an index into the n + 1-gram table.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"A totally non\u00c2\u00ad stochastic rule-based system such as Wang, Li, and Chang's will generally succeed in such cases, but of course runs the risk of overgeneration wherever the single-hanzi word is really intended.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Feature-based HMM Model (Berg- Kirkpatrick et al., 2010): The KM model uses a variety of orthographic features and employs the EM or LBFGS optimization algorithm; Posterior regulariation model (Grac\u00c2\u00b8a et al., 2009): The G10 model uses the posterior regular- ization approach to ensure tag sparsity constraint.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Its only purpose is 3 This follows since each \u00ce\u00b8t has St \u00e2\u0088\u0092 1 parameters and.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The second concerns the methods used (if any) to ex\u00c2\u00ad tend the lexicon beyond the static list of entries provided by the machine-readable dictionary upon which it is based.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Set the decision list to include all rules whose (smoothed) strength is above some threshold Pmin.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In this section, we describe the seven general knowledge sources and explain how the DempsterShafer model makes resolutions.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Once the lexicon has been drawn, the model proceeds similarly to the standard token-level HMM: Emission parameters \u00ce\u00b8 are generated conditioned on tag assignments T . We also draw transition parameters \u00cf\u0086.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We are especially grateful to Taylor Berg- Kirkpatrick for running additional experiments.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"An important reason for separating the two types of features is that this opens up the possibility of theoretical analysis of the use of unlabeled examples.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Considerable amount of work has been done in recent years on the named entity recognition task, partly due to the Message Understanding Conferences (MUC).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"We use a solution to this problem similar to the one presented in (Och et al., 1999), where target words are joined during training.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Thus at each iteration the algorithm is forced to pick features for the location, person and organization in turn for the classifier being trained.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Each distance in the traveling salesman problem now corresponds to the negative logarithm of the product of the translation, alignment and language model probabilities.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Once HMM parameters (\u00ce\u00b8, \u00cf\u0086) are drawn, a token-level tag and word sequence, (t, w), is generated in the standard HMM fashion: a tag sequence t is generated from \u00cf\u0086.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"'Malaysia.'","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The second weakness is purely conceptual, and probably does not affect the per\u00c2\u00ad formance of the model.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"One of the difficulties in Natural Language Processing is the fact that there are many ways to express the same thing or event.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In Table 5 we present results from small test cor\u00c2\u00ad pora for the productive affixes handled by the current version of the system; as with names, the segmentation of morphologically derived words is generally either right or wrong.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The first author is supported by a National Defense Science and Engineering Graduate (NDSEG) fellowship.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"As with personal names, we also derive an estimate from text of the probability of finding a transliterated name of any kind (PTN).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Such constraints are derived from training data, expressing some relationship between features and outcome.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"As with personal names, we also derive an estimate from text of the probability of finding a transliterated name of any kind (PTN).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We have presented a method for unsupervised part- of-speech tagging that considers a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The ATB is disadvantaged by having fewer trees with longer average 5 LDC A-E catalog numbers: LDC2008E61 (ATBp1v4), LDC2008E62 (ATBp2v3), and LDC2008E22 (ATBp3v3.1).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The type-level tag assignments T generate features associated with word types W . The tag assignments constrain the HMM emission parameters \u00ce\u00b8.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"To evaluate the performance on the segmentation task, we report SEG, the standard harmonic means for segmentation Precision and Recall F1 (as defined in Bar-Haim et al. (2005); Tsarfaty (2006)) as well as the segmentation accuracy SEGTok measure indicating the percentage of input tokens assigned the correct exact segmentation (as reported by Cohen and Smith (2007)).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This withdrawal by the treasury secretary is understandable, though.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We consider two variants of Berg-Kirkpatrick et al.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"each word in the lexicon whether or not each string is actually an instance of the word in question.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Another technique for parse hybridization is to use a na\u00efve Bayes classifier to determine which constituents to include in the parse.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"In this paper we study the problem of using a parallel corpus from a background domain (OUT) to improve performance on a target domain (IN) for which a smaller amount of parallel training material\u2014though adequate for reasonable performance\u2014is also available.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"00 76.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Here, $ is the sentence boundary symbol, which is thought to be at position 0 in the target sentence.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The difference in precision between similarity and Bayes switching techniques is significant, but the difference in recall is not.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"47 78.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"None of the models we have presented utilize features associated with a particular constituent (i.e. the label, span, parent label, etc.) to influence parser preference.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Preferences for constituent order (especially in languages with relatively free word order) often belong to this group.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"As a (crude) approximation, we normalize the extraction patterns with respect to active and passive voice and label those extractions as agents or patients.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"These conaUses lossy compression. bThe 8-bit quantized variant returned incorrect probabilities as explained in Section 3.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Our \u201cProjection\u201d baseline is able to benefit from the bilingual information and greatly improves upon the monolingual baselines, but falls short of the \u201cNo LP\u201d model by 2.5% on an average.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The experimental tests are carried out on the Verbmobil task (GermanEnglish, 8000-word vocabulary), which is a limited-domain spoken-language task.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We extend the Stanford parser to accept pre-generated lattices, where each word is represented as a finite state automaton.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Overall, it gives improvements ranging from 1.1% for German to 14.7% for Italian, for an average improvement of 8.3% over the unsupervised feature-HMM model.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Here, annotation proceeds in two phases: first, the domains and the units of IS are marked as such.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"However, those methods need initial seeds, so the relation between entities has to be known in advance.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Moreover, if we want to test on a huge test corpus, indexing the whole corpus might prove computationally expensive.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"With the exception of the Dutch data set, no other processing is performed on the annotated tags.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"This is an iterative method that improves the estimation of the parameters at each iteration.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For example, in .., says Mr. Cooper, a vice president of.. both a spelling feature (that the string contains Mr.) and a contextual feature (that president modifies the string) are strong indications that Mr. Cooper is of type Person.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Roughly speaking, previous work can be divided into three categories, namely purely statistical approaches, purely lexi\u00c2\u00ad cal rule-based approaches, and approaches that combine lexical information with sta\u00c2\u00ad tistical information.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In Figure 4 we show an example of variation between the parsing models.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The model starts by generating a tag assignment for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"4 53.7 43.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"M(wi) = Li).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"This technique has the advantage of requiring no training, but it has the disadvantage of treating all parsers equally even though they may have differing accuracies or may specialize in modeling different phenomena.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In the labeled version of these metrics (L) both heads and arc labels must be correct, while the unlabeled version (U) only considers heads.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"3.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For RandLM, we used the settings in the documentation: 8 bits per value and false positive probability 1 256.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Note that the good classes JADE, GOLD and GRASS have lower costs than the bad classes SICKNESS, DEATH and RAT, as desired, so the trend observed for the results of this method is in the right direction.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"As with lexical expections, the semantic classes of co-referring expressions are 4 They may not be perfectly substitutable, for example one NP may be more specific (e.g., \u00e2\u0080\u009che\u00e2\u0080\u009d vs. \u00e2\u0080\u009cJohn F. Kennedy\u00e2\u0080\u009d).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The work of the first author was supported by the Lynn and William Frankel Center for Computer Sciences.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"For simplicity, we assume that OUT is homogeneous.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In practice, we can therefore expect a trade-off such that increasing the amount of information encoded in arc labels will cause an increase in the accuracy of the inverse transformation but a decrease in the accuracy with which the parser can construct the labeled representations.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Daum\u00b4e (2007) applies a related idea in a simpler way, by splitting features into general and domain-specific versions.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"This departure from the traditional token-based tagging approach allows us to explicitly capture type- level distributional properties of valid POS tag as signments as part of the model.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"16 As one reviewer points out, one problem with the unigram model chosen here is that there is still a. tendency to pick a segmentation containing fewer words.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Purely statistical approaches have not been very popular, and so far as we are aware earlier work by Sproat and Shih (1990) is the only published instance of such an approach.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We have described grammar state splits that significantly improve parsing performance, catalogued parsing errors, and quantified the effect of segmentation errors.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"For example, a good classifier would identify Mrs. Frank as a person, Steptoe & Johnson as a company, and Honduras as a location.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Since Dane\u00cb\u0087s\u00e2\u0080\u0099 proposals of \u00e2\u0080\u0098thematic development patterns\u00e2\u0080\u0099, a few suggestions have been made as to the existence of a level of discourse structure that would predict the information structure of sentences within texts.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Month Names, Days of the Week, and Numbers: If is initCaps and is one of January, February, . . .","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"(2009b) evaluated the Bikel parser using the same ATB split, but only reported dev set results with gold POS tags for sentences of length \u00e2\u0089\u00a4 40.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"A corpus of German newspaper commentaries has been assembled at Potsdam University, and annotated with different linguistic information, to different degrees.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"na me =>1 ha nzi fa mi ly 1 ha nzi gi ve n 4.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"we perform five runs with different random initialization of sampling state.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We therefore used the arithmetic mean of each interjudge precision-recall pair as a single measure of interjudge similarity.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Another option is the closedsource data structures from Sheffield (Guthrie and Hepple, 2010).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"A cell in the bottom row of the parse chart is required for each potential whitespace boundary.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In contrast, even though context can be helpful for resolving definite NPs, context can be trumped by the semantics of the nouns themselves.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"It is known that CFG's, HG's, and TAG's can be recognized in polynomial time since polynomial time algorithms exist in for each of these formalisms.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"For instance, on Spanish, the absolute gap on median performance is 10%.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Acknowledgments We thank Meni Adler and Michael Elhadad (BGU) for helpful comments and discussion.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We have described two data structures for language modeling that achieve substantial reductions in time and memory cost.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If it starts with a lower case letter, and contains both upper and lower case letters, then (mixedCaps, zone) is set to 1.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"While the first three models get three to four tags wrong, our best model gets only one word wrong and is the most accurate among the four models for this example.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The difference is due to more careful S-NOM NP NP NP VP VBG :: b NP restoring NP ADJP NN :: b NP NN NP NP ADJP DTJJ ADJP DTJJ NN :: b NP NP NP ADJP ADJP DTJJ J ..i NN :: b NP NP NP ADJP ADJP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ DTJJ J ..i _;\u00ef\u00bf\u00bd PRP J ..i NN _;\u00ef\u00bf\u00bd NP PRP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ J ..i role its constructive effective (b) Stanford (c) Berkeley (d) Bik el (a) Reference Figure 4: The constituent Restoring of its constructive and effective role parsed by the three different models (gold segmentation).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The similar explanation applies to the link to the \u00e2\u0080\u009cstake\u00e2\u0080\u009d set.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"92 77.","label":0},{"source_text":"There is no global pruning.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Find keywords for each NE pair The keywords are found for each NE category pair.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"A simple lexicalized PCFG with second order Markovization gives relatively poor performance: 75.95% F1 on the test set.8 But this figure is surprisingly competitive with a recent state-of-the-art baseline (Table 7).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Entries landing in the same bucket are said to collide.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"For each 13 Of course, this weighting makes the PCFG an improper distribution.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Figure 2: Order in which source positions are visited for the example given in Fig.1.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Base NPs are the other significant category of nominal phrases.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"This group consists of (3 total number of possible zones) features.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Table 3: Dev set frequencies for the two most significant discourse markers in Arabic are skewed toward analysis as a conjunction.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In total, for the 2,000 NE category pairs, 5,184 keywords are found.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Only one of the first n positions which are not already aligned in a partial hypothesis may be chosen, where n is set to 4.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Several systems propose statistical methods for handling unknown words (Chang et al. 1992; Lin, Chiang, and Su 1993; Peng and Chang 1993).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This is an issue that we have not addressed at the current stage of our research.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Unlike Kahane et al. (1998), we do not regard a projectivized representation as the final target of the parsing process.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The parallel data came from the Europarl corpus (Koehn, 2005) and the ODS United Nations dataset (UN, 2006).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The compressed variant uses block compression and is rather slow as a result.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We call such a constituent an isolated constituent.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"\u00e2\u0080\u0094ested in correlations between prosody and dis course structure.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"2 70.7 52.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The ATB gives several different analyses to these words to indicate different types of coordination.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"LCFRS's have only been loosely defined in this paper; we have yet to provide a complete set of formal properties associated with members of this class.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The label for a test example with features x is then defined as In this paper we define h(x, y) as the following function of counts seen in training data: Count(x,y) is the number of times feature x is seen with label y in training data, Count(x) = EyEy Count(x, y). a is a smoothing parameter, and k is the number of possible labels.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The paper is structured as follows.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Context from the whole document can be important in classifying a named entity.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"In general, the lemma of the previous section does not ensure that all the productions in the combined parse are found in the grammars of the member parsers.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Words and punctuation that appear in brackets are considered optional.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The features are weighted within a logistic model to give an overall weight that is applied to the phrase pair\u2019s frequency prior to making MAP-smoothed relative-frequency estimates (different weights are learned for each conditioning direction).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"To approximate these baselines, we implemented a very simple sentence selection algorithm in which parallel sentence pairs from OUT are ranked by the perplexity of their target half according to the IN language model.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This fully generative model caters for real interaction between the syntactic and morphological levels as a part of a single coherent process.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"phrase (markContainsVerb).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Variants of alif are inconsistently used in Arabic texts.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"2.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The major problem for all segmentation systems remains the coverage afforded by the dictionary and the lexical rules used to augment the dictionary to deal with unseen words.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Table 1 The cost as a novel given name (second position) for hanzi from various radical classes.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For instance, the common \"suffixes,\" -nia (e.g.,.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Given counts cn1 where e.g. c1 is the vocabulary size, total memory consumption, in bits, is Our PROBING data structure places all n-grams of the same order into a single giant hash table.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"This paper is based on work supported in part by DARPA through IBM.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"These alignment models are similar to the concept of hidden Markov models (HMM) in speech recognition.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"rhetorical analysis We are experimenting with a hybrid statistical and knowledge-based system for discourse parsing and summarization (Stede 2003), (Hanneforth et al. 2003), again targeting the genre of commentaries.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Further, it needs extra pointers in the trie, increasing model size by 40%.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Even with vocalization, there are linguistic categories that are difficult to identify without semantic clues.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Label propagation is used to propagate these tags inwards and results in tag distributions for the middle word of each Italian trigram.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"com \u00c2\u00a7Cambridge, UK Email: nc201@eng.cam.ac.uk \u00c2\u00a9 1996 Association for Computational Linguistics (a) B ) ( , : & ; ? ' H o w d o y o u s a y o c t o p u s i n J a p a n e s e ? ' (b) P l a u s i b l e S e g m e n t a t i o n I B X I I 1 : & I 0 0 r i 4 w e n 2 z h a n g l y u 2 z e n 3 m e 0 s h u o l ' J a p a n e s e ' ' o c t o p u s ' ' h o w ' ' s a y ' (c) Figure 1 I m p l a u s i b l e S e g m e n t a t i o n [\u00c2\u00a7] lxI 1:&I ri4 wen2 zhangl yu2zen3 me0 shuol 'Japan' 'essay' 'fish' 'how' 'say' A Chinese sentence in (a) illustrating the lack of word boundaries.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The combining technique must act as a multi-position switch indicating which parser should be trusted for the particular sentence.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If the token is the first word of a sentence, then this feature is set to 1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The cost is computed as follows, where N is the corpus size and f is the frequency: (1) Besides actual words from the base dictionary, the lexicon contains all hanzi in the Big 5 Chinese code\/ with their pronunciation(s), plus entries for other characters that can be found in Chinese text, such as Roman letters, numerals, and special symbols.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Step 3.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"irL as the product of the probability estimate for i\u00c2\u00a5JJ1l., and the probability estimate just derived for unseen plurals in ir,: p(i\u00c2\u00a51J1l.ir,) p(i\u00c2\u00a51J1l.)p(unseen(f,)).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Using these states, we define partial hypothesis extensions, which are of the following type: (S0;C n fjg; j0) !","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"2.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Not all the layers have been produced for all the texts yet.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"(3)In sentence (1), McCann can be a person or an orga nization.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The result of this is shown in Figure 7.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Previous work has attempted to incorporate such constraints into token-level models via heavy-handed modifications to inference procedure and objective function (e.g., posterior regularization and ILP decoding) (Grac\u00c2\u00b8a et al., 2009; Ravi and Knight, 2009).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The PROBING model is 2.4 times as fast as the fastest alternative, SRILM, and uses less memory too.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Recall.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"As mentioned above, it is not obvious how to apply Daum\u00b4e\u2019s approach to multinomials, which do not have a mechanism for combining split features.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The CFLex and CFNet knowledge sources provide positive evidence that a candidate NP and anaphor might be coreferent.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This is a straightforward technique that is arguably better suited to the adaptation task than the standard method of treating representative IN sentences as queries, then pooling the match results.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This is appropriate in cases where it is sanctioned by Bayes\u2019 law, such as multiplying LM and TM probabilities, but for adaptation a more suitable framework is often a mixture model in which each event may be generated from some domain.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Links can be of two different kinds: anaphoric or bridging (definite noun phrases picking up an antecedent via world-knowledge).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Two aspects of the corpus have been presented in previous papers ((Re- itter, Stede 2003) on underspecified rhetorical structure; (Stede 2003) on the perspective of knowledge-based summarization).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Nonetheless, the prospect of a network of annotated discourse resources seems particularly promising if not only a single annotation layer is used but a whole variety of them, so that a systematic search for correlations between them becomes possible, which in turn can lead to more explanatory models of discourse structure.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"If is a number string (such as one, two, etc), then the feature NumberString is set to 1.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"While processing the source sentence monotonically, the initial state I is entered whenever there are no uncovered positions to the left of the rightmost covered position.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The Bayes models were able to achieve significantly higher precision than their non-parametric counterparts.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Thus, the effects of spontaneous speech are present in the corpus, e.g. the syntactic structure of the sentence is rather less restricted, however the effect of speech recognition errors is not covered.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Global features are extracted from other occurrences of the same token in the whole document.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kind of cue.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For a given partial hypothesis (C; j), the order in which the cities in C have been visited can be ignored (except j), only the score for the best path reaching j has to be stored.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"However, using the top-level semantic classes of WordNet proved to be problematic because the class distinctions are too coarse.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Specifically, for the ith word type, the set of token-level tags associated with token occurrences of this word, denoted t(i), must all take the value Ti to have nonzero mass. Thus in the context of Gibbs sampling, if we want to block sample Ti with t(i), we only need sample values for Ti and consider this setting of t(i).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Gather phrases using keywords Now, the keyword with the top TF\/ITF score is selected for each phrase.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The first modification \u2014 cautiousness \u2014 is a relatively minor change.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In the cotraining case, (Blum and Mitchell 98) argue that the task should be to induce functions Ii and f2 such that So Ii and 12 must (1) correctly classify the labeled examples, and (2) must agree with each other on the unlabeled examples.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The semantic caseframe expectations are used in two ways.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The equation for sampling a single type-level assignment Ti is given by, 0.2 0 5 10 15 20 25 30 Iteration Figure 2: Graph of the one-to-one accuracy of our full model (+FEATS) under the best hyperparameter setting by iteration (see Section 5).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Inflectional features marking pronominal elements may be attached to different kinds of categories marking their pronominal complements.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We have shown that, at least given independent human judgments, this is not the case, and that therefore such simplistic measures should be mistrusted.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"SRILM (Stolcke, 2002) is widely used within academia.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For each domain, we created a semantic dictionary by doing two things.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our motivation for using DempsterShafer is that it provides a well-principled framework for combining evidence from multiple sources with respect to competing hypotheses.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Since the transducers are built from human-readable descriptions using a lexical toolkit (Sproat 1995), the system is easily maintained and extended.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"It has been shown in (Chandra et al., 1981) that if M works in space log n then there is a deterministic TM which accepts the same language in polynomial time.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"As in boosting, the algorithm works in rounds.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Dynamic programming efficiently scores many hypotheses by exploiting the fact that an N-gram language model conditions on at most N \u2212 1 preceding words.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We would also like to thank Amarnag Subramanya for helping us with the implementation of label propagation and Shankar Kumar for access to the parallel data.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Table 4 shows how much the Bayes switching technique uses each of the parsers on the test set.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Words and punctuation that appear in brackets are considered optional.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The approach uses both spelling and contextual rules.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Experiments using data from the Prague Dependency Treebank show that the combined system can handle nonprojective constructions with a precision sufficient to yield a significant improvement in overall parsing accuracy.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"It has no syntactic function.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Our new approach, which is called quasi-monotone search, processes the source sentence monotonically, while explicitly taking into account the positions of the German verbgroup.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Quantization can be improved by jointly encoding probability and backoff.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Some of these approaches (e.g., Lin, Chiang, and Su [1993]) attempt to identify unknown words, but do not ac\u00c2\u00ad tually tag the words as belonging to one or another class of expression.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Here, the pruning threshold t0 = 10:0 is used.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"For all these annotation taks, G\u00c2\u00a8otze developed a series of questions (essentially a decision tree) designed to lead the annotator to the ap propriate judgement.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In addition to the tapes required to store the indices, M requires one work tape for splitting the substrings.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"During testing, it is possible that the classifier produces a sequence of inadmissible classes (e.g., person begin followed by location unique).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Two measures that can be used to compare judgments are: 1.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Note that in our model, conditioned on T , there is precisely one t which has nonzero probability for the token component, since for each word, exactly one \u00ce\u00b8t has support.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In the second part of the experiment, we applied the inverse transformation based on breadth-first search under the three different encoding schemes.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Obviously \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is part of the following clause rather than being the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d and the extracted instance makes no sense.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Skipped (K): The translation of up to one word may be postponed . Verb (V): The translation of up to two words may be anticipated.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"4 70.4 46.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"These performance gains transfer to improved system runtime performance; though we focused on Moses, our code is the best lossless option with cdec and Joshua.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Much confusion has been sown about Chinese writing by the use of the term ideograph, suggesting that hanzi somehow directly represent ideas.","label":0},{"source_text":"There is no global pruning.","target_text":"In this case, we have no finite-state restrictions for the search space.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"We can do that . Input: Das ist zu knapp , weil ich ab dem dritten in Kaiserslautern bin . Genaugenommen nur am dritten . Wie ware es denn am ahm Samstag , dem zehnten Februar ? MonS: That is too tight , because I from the third in Kaiserslautern . In fact only on the third . How about ahm Saturday , the tenth of February ? QmS: That is too tight , because I am from the third in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? IbmS: That is too tight , from the third because I will be in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? Input: Wenn Sie dann noch den siebzehnten konnten , ware das toll , ja . MonS: If you then also the seventeenth could , would be the great , yes . QmS: If you could then also the seventeenth , that would be great , yes . IbmS: Then if you could even take seventeenth , that would be great , yes . Input: Ja , das kommt mir sehr gelegen . Machen wir es dann am besten so . MonS: Yes , that suits me perfectly . Do we should best like that . QmS: Yes , that suits me fine . We do it like that then best . IbmS: Yes , that suits me fine . We should best do it like that .","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"A crucial difference is that the number of parameters is greatly reduced as is the number of variables that are sampled during each iteration.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"When finished, the whole material is written into an XML-structured annotation file.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"All improvements over the baseline are statistically significant beyond the 0.01 level (McNemar\u2019s test).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Figure 3 shows examples of semantic expectations that were learned.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The algorithm in Fig.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We found that contextual role knowledge was more beneficial for pronouns than for definite noun phrases.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"They cluster NE instance pairs based on the words in the contexts using a bag- of-words method.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In application settings, this may be a profitable strategy.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Syntactic universals are a well studied concept in linguistics (Carnie, 2002; Newmeyer, 2005), and were recently used in similar form by Naseem et al. (2010) for multilingual grammar induction.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Personal names such as 00, 3R; zhoulenl-lai2 'Zhou Enlai.'","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Our model outperforms theirs on four out of five languages on the best hyperparameter setting and three out of five on the median setting, yielding an average absolute difference across languages of 12.9% and 3.9% for best and median settings respectively compared to their best EM or LBFGS performance.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"If the expression is a word or a short phrase (like \u00e2\u0080\u009ccorporation\u00e2\u0080\u009d and \u00e2\u0080\u009ccompany\u00e2\u0080\u009d), it is called a \u00e2\u0080\u009csynonym\u00e2\u0080\u009d.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Each knowledge source then assigns a probability estimate to each candidate, which represents its belief that the candidate is the antecedent for the anaphor.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"As can be seen from the last column in Table 1, both Head and Head+Path may theoretically lead to a quadratic increase in the number of distinct arc labels (Head+Path being worse than Head only by a constant factor), while the increase is only linear in the case of Path.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"0 X u} \"' o; .2 X X><X X XX X X X X X X x X X X X X x X V X X X X .;t'*- XXX:OX X X X X X X 9 x X X XX XX X X X X X X X XXX:< X X>O<XX>!KXX XI<>< \u00c2\u00bbC X X XX :X: X X \"' X X XX >OO<X>D<XIK X X X X X X --XX\u00c2\u00bb: XXX X X\u00c2\u00bbC X X\u00c2\u00abX...C:XXX X Xll< X X ><XX>IIC:liiC:oiiiiCI--8!X:liiOC!I!S8K X X X 10 100 1000 10000 log(F)_base: R\"2=0.20 (p < 0.005) X 100000 Figure 6 Plot of log frequency of base noun, against log frequency of plural nouns.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"0 70.9 42.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In a few cases, the criteria for correctness are made more explicit.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"(1998) have also used a maximum entropy classifier that uses already tagged entities to help tag other entities.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Future work along these lines will incorporate other layers of annotation, in particular the syntax information.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Cohen and Smith (2007) later on based a system for joint inference on factored, independent, morphological and syntactic components of which scores are combined to cater for the joint inference task.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"rently, some annotations (in particular the connectives and scopes) have already moved beyond the core corpus; the others will grow step by step.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The Berkeley parser gives state-of-the-art performance for all metrics.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Whether a language even has orthographic words is largely dependent on the writing system used to represent the language (rather than the language itself); the notion \"orthographic word\" is not universal.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Two subjects are each given a calendar and they are asked to schedule a meeting.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(2) was extended to have an additional, innermost loop over the (3) possible labels.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The token-level term is similar to the standard HMM sampling equations found in Johnson (2007).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"As noted in Section 4.4, disk cache state is controlled by reading the entire binary file before each test begins.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In addition to the named-entity string (Maury Cooper or Georgia), a contextual predictor was also extracted.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"To measure the contribution of each modification, a third, intermediate algorithm, Yarowsky-cautious was also tested.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In the ATB, :: b asta\u00e2\u0080\u0099adah is tagged 48 times as a noun and 9 times as verbal noun.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(1992).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Tsarfaty (2006) argues that for Semitic languages determining the correct morphological segmentation is dependent on syntactic context and shows that increasing information sharing between the morphological and the syntactic components leads to improved performance on the joint task.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Their default variant implements a forward trie, in which words are looked up in their natural left-to-right order.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We also mark all nodes that dominate an SVO configuration (containsSVO).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"a classifier.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"rhetorical analysis We are experimenting with a hybrid statistical and knowledge-based system for discourse parsing and summarization (Stede 2003), (Hanneforth et al. 2003), again targeting the genre of commentaries.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In general, m; l; l0 6= fl1; l2; l3g and in line umber 3 and 4, l0 must be chosen not to violate the above reordering restriction.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We will evaluate various specific aspects of the segmentation, as well as the overall segmentation per\u00c2\u00ad formance.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"There are still some open issues to be resolved with the format, but it represents a first step.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Ex: Mr. Cristiani, president of the country ...","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"07 95.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The method shares some characteristics of the decision list algorithm presented in this paper.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The entries in this table can be compared with those of Table 3 to see how the performance of the combining techniques degrades in the presence of an inferior parser.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The focus of our work is on the use of contextual role knowledge for coreference resolution.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Dictionaries: Due to the limited amount of training material, name dictionaries have been found to be useful in the named entity task.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Acronyms (ACRO): Words made up of all capitalized letters in the text zone will be stored as acronyms (e.g., IBM).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We consider two variants of Berg-Kirkpatrick et al.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Compared with the widely- SRILM, our is 2.4 times as fast while using 57% of the mem- The structure is a trie with bit-level packing, sorted records, interpolation search, and optional quantization aimed lower memory consumption. simultaneously uses less memory than the smallest lossless baseline and less CPU than the baseline.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"In addition to the optimizations specific to each datastructure described in Section 2, we implement several general optimizations for language modeling.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"BABAR uses the log-likelihood statistic (Dunning, 1993) to evaluate the strength of a co-occurrence relationship.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Raj and Whittaker (2003) show that integers in a trie implementation can be compressed substantially.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Thus, we feel fairly confident that for the examples we have considered from Gan's study a solution can be incorporated, or at least approximated, within a finite-state framework.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The weak hypothesis can abstain from predicting the label of an instance x by setting h(x) = 0.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Even without features, but still using the tag prior, our median result is 52.0%, still significantly outperforming Grac\u00c2\u00b8a et al.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"First, we parsed the training corpus, collected all the noun phrases, and looked up each head noun in WordNet (Miller, 1990).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Aside from adding a simple rule to correct alif deletion caused by the preposition J, no other language-specific processing is performed.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"1 61.2 43.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For eight judges, ranging k between 1 and 8 corresponded to a precision score range of 90% to 30%, meaning that there were relatively few words (30% of those found by the automatic segmenter) on which all judges agreed, whereas most of the words found by the segmenter were such that one human judge agreed.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Among machine learning-based NERs, Identi- Finder has proven to be the best on the official MUC6 and MUC7 test data.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Table 2 shows these similarity measures.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We have not explored this strategy.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"With regard to purely morphological phenomena, certain processes are not han\u00c2\u00ad dled elegantly within the current framework Any process involving reduplication, for instance, does not lend itself to modeling by finite-state techniques, since there is no way that finite-state networks can directly implement the copying operations required.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"If the expression is a word or a short phrase (like \u00e2\u0080\u009ccorporation\u00e2\u0080\u009d and \u00e2\u0080\u009ccompany\u00e2\u0080\u009d), it is called a \u00e2\u0080\u009csynonym\u00e2\u0080\u009d.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Both parameters depend on a single hyperparameter \u00ce\u00b1.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Pseudo-Projective Dependency Parsing","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Taking the intersection of languages in these resources, and selecting languages with large amounts of parallel data, yields the following set of eight Indo-European languages: Danish, Dutch, German, Greek, Italian, Portuguese, Spanish and Swedish.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"To provide a thorough analysis, we evaluated three baselines and two oracles in addition to two variants of our graph-based approach.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"But we follow the more direct adaptation of Evalb suggested by Tsarfaty (2006), who viewed exact segmentation as the ultimate goal.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This significantly underperforms log-linear combination.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Sometimes extracted phrases by themselves are not meaningful to consider without context, but we set the following criteria.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We indicate whether a context with zero log backoff will extend using the sign bit: +0.0 for contexts that extend and \u22120.0 for contexts that do not extend.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In this case, we have no finite-state restrictions for the search space.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"F-measure is the harmonic mean of precision and recall, 2PR\/(P + R).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"(2006).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Clearly, explicitly modeling such a powerful constraint on tagging assignment has a potential to significantly improve the accuracy of an unsupervised part-of-speech tagger learned without a tagging dictionary.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Table 4 shows the results.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The NE tagger is a rule-based system with 140 NE categories [Sekine et al. 2004].","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Although this is not a precise criterion, most cases we evaluated were relatively clear-cut.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Methods for expanding the dictionary include, of course, morphological rules, rules for segmenting personal names, as well as numeral sequences, expressions for dates, and so forth (Chen and Liu 1992; Wang, Li, and Chang 1992; Chang and Chen 1993; Nie, Jin, and Hannan 1994).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(4) gives In order to minimize Zt, at each iteration the final algorithm should choose the weak hypothesis (i.e., a feature xt) which has values for W+ and W_ that minimize Equ.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Taking \/V, to be the number of examples an algorithm classified correctly (where all gold standard items labeled noise were counted as being incorrect), we calculated two measures of accuracy: See Tab.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"For each language and setting, we report one-to-one (11) and many- to-one (m-1) accuracies.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"2.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Lexical and Morphological Ambiguity The rich morphological processes for deriving Hebrew stems give rise to a high degree of ambiguity for Hebrew space-delimited tokens.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, the passive voice pattern \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d and the active voice pattern \u00e2\u0080\u009ckidnapped <direct object>\u00e2\u0080\u009d are merged into a single normalized pattern \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d.2 For the sake of sim plicity, we will refer to these normalized extraction patterns as caseframes.3 These caseframes can capture two types of contextual role information: (1) thematic roles corresponding to events (e.g, \u00e2\u0080\u009c<agent> kidnapped\u00e2\u0080\u009d or \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d), and (2) predicate-argument relations associated with both verbs and nouns (e.g., \u00e2\u0080\u009ckidnapped for <np>\u00e2\u0080\u009d or \u00e2\u0080\u009cvehicle with <np>\u00e2\u0080\u009d).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Combining multiple highly-accurate independent parsers yields promising results.","label":1},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Adding the isolated constituents to our hypothesis parse could increase our expected recall, but in the cases we investigated it would invariably hurt our precision more than we would gain on recall.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The set of candidate constituents comes from the union of all the constituents suggested by the member parsers.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"attaching to terms denoting human beings.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"P St = n. \u00ce\u00b2 T VARIABLES \u00cf\u0088 Y W : Word types (W1 ,.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"att.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The rest of the paper is structured as follows.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Section 3 describes the complete coreference resolution model, which uses the contextual role knowledge as well as more traditional coreference features.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Notice that the CC-domain is a special case.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"3 These are not full case frames in the traditional sense, but they approximate a simple case frame with a single slot.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The following features were used: full-string=x The full string (e.g., for Maury Cooper, full- s tring=Maury_Cooper). contains(x) If the spelling contains more than one word, this feature applies for any words that the string contains (e.g., Maury Cooper contributes two such features, contains (Maury) and contains (Cooper) . allcapl This feature appears if the spelling is a single word which is all capitals (e.g., IBM would contribute this feature). allcap2 This feature appears if the spelling is a single word which is all capitals or full periods, and contains at least one period.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"As the two NE categories are the same, we can\u00e2\u0080\u0099t differentiate phrases with different orders of par ticipants \u00e2\u0080\u0093 whether the buying company or the to-be-bought company comes first.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Although the ultimate goal of the Verbmobil project is the translation of spoken language, the input used for the translation experiments reported on in this paper is the (more or less) correct orthographic transcription of the spoken sentences.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Some relations are signalled by subordinating conjunctions, which clearly demarcate the range of the text spans related (matrix clause, embedded clause).","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In approaching this problem, a variety of different methods are conceivable, including a more or less sophisticated use of machine learning.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"One possibility is to use n-grams based on mutual information.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Clearly, explicitly modeling such a powerful constraint on tagging assignment has a potential to significantly improve the accuracy of an unsupervised part-of-speech tagger learned without a tagging dictionary.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"A related point is that mutual information is helpful in augmenting existing electronic dictionaries, (cf.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Unlike Kahane et al. (1998), we do not regard a projectivized representation as the final target of the parsing process.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Nonetheless, parse quality is much lower in the joint model because a lattice is effectively a long sentence.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"A very small excerpt from an Italian-English graph is shown in Figure 1.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Among machine learning-based NERs, Identi- Finder has proven to be the best on the official MUC6 and MUC7 test data.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For even larger models, storing counts (Talbot and Osborne, 2007; Pauls and Klein, 2011; Guthrie and Hepple, 2010) is a possibility.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While building a machine translation system is a serious undertaking, in future we hope to attract more newcomers to the field by keeping the barrier of entry as low as possible.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"However, since we want to preserve as much of the original structure as possible, we are interested in finding a transformation that involves a minimal number of lifts.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Given names are most commonly two hanzi long, occasionally one hanzi long: there are thus four possible name types, which can be described by a simple set of context-free rewrite rules such as the following: 1.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"A typical situation is shown in Figure 1.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The data structure was populated with 64-bit integers sampled uniformly without replacement.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In focus is in particular the correlation with rhetorical structure, i.e., the question whether specific rhetorical relations \u00e2\u0080\u0094 or groups of relations in particular configurations \u00e2\u0080\u0094 are signalled by speakers with prosodic means.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"6 Joint Segmentation and Parsing.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"However, when we pre- tag the input\u00e2\u0080\u0094as is recommended for English\u00e2\u0080\u0094 we notice a 0.57% F1 improvement.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The same numbers were used for each data structure.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Juri Ganitkevitch answered questions about Joshua.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"A total of 13,976 phrases were grouped.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Word type N % Dic tion ary entr ies 2 , 5 4 3 9 7 . 4 7 Mor pho logi call y deri ved wor ds 3 0 . 1 1 Fore ign tran slite rati ons 9 0 . 3 4 Per son al na mes 5 4 2 . 0 7 cases.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"A simple lexicalized PCFG with second order Markovization gives relatively poor performance: 75.95% F1 on the test set.8 But this figure is surprisingly competitive with a recent state-of-the-art baseline (Table 7).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"However, supervised methods rely on labeled training data, which is time-consuming and expensive to generate.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The number of ATB n-grams also falls below the WSJ sample size as the largest WSJ sample appeared in only 162 corpus positions.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Global context from the whole document is available and can be exploited in a natural manner with a maximum entropy classifier.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Then, token- level HMM emission parameters are drawn conditioned on these assignments such that each word is only allowed probability mass on a single assigned tag.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We present KenLM, a library that implements two data structures for efficient language model queries, reducing both time and costs.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"We developed a first version of annotation guidelines for co-reference in PCC (Gross 2003), which served as basis for annotating the core corpus but have not been empirically evaluated for inter-annotator agreement yet.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"8 57.3 +F EA TS be st me dia n 50.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Otherwise, the scope of the search problem shrinks recursively: if A[pivot] < k then this becomes the new lower bound: l +\u2014 pivot; if A[pivot] > k then u +\u2014 pivot.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Asking the annotator to also formulate the question is a way of arriving at more reproducible decisions.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"8 1 8.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"0 70.9 42.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"We assume that the goal in dependency parsing is to construct a labeled dependency graph of the kind depicted in Figure 1.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We believe it is natural for authors to use abbreviations in subsequent mentions of a named entity (i.e., first \u00e2\u0080\u009cPresident George Bush\u00e2\u0080\u009d then \u00e2\u0080\u009cBush\u00e2\u0080\u009d).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The machine learning community has been in a similar situation and has studied the combination of multiple classifiers (Wolpert, 1992; Heath et al., 1996).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Figure 3 shows a small fragment of the WFST encoding the dictionary, containing both entries forjust discussed, g:t\u00c2\u00a5 zhonglhua2 min2guo2 (China Republic) 'Republic of China,' and i\u00c2\u00a5inl.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"To eliminate such sequences, we define a transition probability between word classes to be equal to 1 if the sequence is admissible, and 0 otherwise.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"PoS tags impose a unique morphological segmentation on surface tokens and present a unique valid yield for syntactic trees.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Manual annotation results in human in- terpretable grammars that can inform future tree- bank annotation decisions.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"TAG's can be shown to be equivalent to this restricted system.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"an event.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"If the expression is longer or complicated (like \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d and \u00e2\u0080\u009cA\u00e2\u0080\u0099s purchase of B\u00e2\u0080\u009d), it is called \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d, i.e. a set of phrases which express the same thing or event.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Our analysis and comparison focuses primarily on the one-to-one accuracy since it is a stricter metric than many-to-one accuracy, but also report many-to-one for completeness.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The second setting uses the news-related subcorpora for the NIST09 MT Chinese to English evaluation8 as IN, and the remaining NIST parallel Chinese\/English corpora (UN, Hong Kong Laws, and Hong Kong Hansard) as OUT.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Mi(c) is a binary function returning t when parser i (from among the k parsers) suggests constituent c should be in the parse.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The precision and recall of similarity switching and constituent voting are both significantly better than the best individual parser, and constituent voting is significantly better than parser switching in precision.4 Constituent voting gives the highest accuracy for parsing the Penn Treebank reported to date.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"However, the learning curves in Figure 3 show that the Berkeley parser does not exceed our manual grammar by as wide a margin as has been shown for other languages (Petrov, 2009).","label":0},{"source_text":"There is no global pruning.","target_text":"The details are given in (Och and Ney, 2000).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"3 68.9 50.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"To combat the proliferation of parsing edges, we prune the lattices according to a hand-constructed lexicon of 31 clitics listed in the ATB annotation guidelines (Maamouri et al., 2009a).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The choice between different actions is in general nondeterministic, and the parser relies on a memorybased classifier, trained on treebank data, to predict the next action based on features of the current parser configuration.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Coverage indicates the fraction of hypotheses in which the character yield exactly matched the reference.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Since the inclusion of out-ofdomain test data was a very late decision, the participants were not informed of this.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"A dynamic programming algorithm is then used to select the sequence of word classes with the highest probability.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"As with lexical expections, the semantic classes of co-referring expressions are 4 They may not be perfectly substitutable, for example one NP may be more specific (e.g., \u00e2\u0080\u009che\u00e2\u0080\u009d vs. \u00e2\u0080\u009cJohn F. Kennedy\u00e2\u0080\u009d).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Taking \/V, to be the number of examples an algorithm classified correctly (where all gold standard items labeled noise were counted as being incorrect), we calculated two measures of accuracy: See Tab.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Evaluation We use 8 different measures to evaluate the performance of our system on the joint disambiguation task.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"In these experiments, the input lacks segmentation markers, hence the slightly different dev set baseline than in Table 6.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"constitute names, since we have only their segmentation, not the actual classification of the segmented words.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Cohen and Smith (2007) chose a metric like SParseval (Roark et al., 2006) that first aligns the trees and then penalizes segmentation errors with an edit-distance metric.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"When extracting the vector t, used to compute the constraint feature from the graph, we tried three threshold values for r (see Eq.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Fortunately, we were able to obtain a copy of the full set of sentences from Chang et al. on which Wang, Li, and Chang tested their system, along with the output of their system.18 In what follows we will discuss all cases from this set where our performance on names differs from that of Wang, Li, and Chang.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A simple extension will be used to handle this problem.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The general-language features have a slight advantage over the similarity features, and both are better than the SVM feature.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"30 75.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"JI!","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"One may argue with these efforts on normalization, and ultimately their value should be assessed by assessing their impact on inter-annotator agreement.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"na me =>2 ha nzi fa mi ly 2 ha nzi gi ve n 5.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Twentieth-century linguistic work on Chinese (Chao 1968; Li and Thompson 1981; Tang 1988,1989, inter alia) has revealed the incorrectness of this traditional view.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The results clearly indicate increased variation in the ATB relative to the WSJ, but care should be taken in assessing the magnitude of the difference.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In total 13,976 phrases are assigned to sets of phrases, and the accuracy on our evaluation data ranges from 65 to 99%, depending on the domain and the size of the sets.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"In graph-based learning approaches one constructs a graph whose vertices are labeled and unlabeled examples, and whose weighted edges encode the degree to which the examples they link have the same label (Zhu et al., 2003).","label":0},{"source_text":"Here we present two algorithms.","target_text":"AdaBoost was first introduced in (Freund and Schapire 97); (Schapire and Singer 98) gave a generalization of AdaBoost which we will use in this paper.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The data structure was populated with 64-bit integers sampled uniformly without replacement.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"so that 'door' would be and in this case the hanzi 7C, does not represent a syllable.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Search t0 CPU time #search mWER Method [sec] error [%] QmS 0.0 0.07 108 42:6 1.0 0.13 85 37:8 2.5 0.35 44 36:6 5.0 1.92 4 34:6 10.0 10.6 0 34:5 IbmS 0.0 0.14 108 43:4 1.0 0.3 84 39:5 2.5 0.8 45 39:1 5.0 4.99 7 38:3 10.0 28.52 0 38:2 Table 6 shows example translations obtained by the three different approaches.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Exposing this information to the decoder will lead to better hypothesis recombination.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Figure 2 shows timing results.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"37 79.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"At first glance, this seems only peripherally related to our work, since the specific\/general distinction is made for features rather than instances.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The PROBING model can perform optimistic searches by jumping to any n-gram without needing state and without any additional memory.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We can better predict the probability of an unseen hanzi occurring in a name by computing a within-class Good-Turing estimate for each radical class.","label":0},{"source_text":"There is no global pruning.","target_text":"Note that in line 4 the last visited position for the successor hypothesis must be m. Otherwise , there will be four uncovered positions for the predecessor hypothesis violating the restriction.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Obviously \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is part of the following clause rather than being the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d and the extracted instance makes no sense.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In the namedentity problem each example is a (spelling,context) pair.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Table 4 shows translation results for the three approaches.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Training examples are generated automatically by identifying noun phrases that can be easily resolved with their antecedents using lexical and syntactic heuristics.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The normalization factor plays an important role in the AdaBoost algorithm.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The second algorithm extends ideas from boosting algorithms, designed for supervised learning tasks, to the framework suggested by (Blum and Mitchell 98).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"In (Reitter, Stede 2003) we went a different way and suggested URML5, an XML format for underspecifying rhetorical structure: a number of relations can be assigned instead of a single one, competing analyses can be represented with shared forests.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Ignoring the identity of the target language words e and e0, the possible partial hypothesis extensions due to the IBM restrictions are shown in Table 2.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"In our case multi-threading is trivial because our data structures are read-only and uncached.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Another approach to finding paraphrases is to find phrases which take similar subjects and objects in large corpora by using mutual information of word distribution [Lin and Pantel 01].","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Search CPU time mWER SSER Method [sec] [%] [%] MonS 0:9 42:0 30:5 QmS 10:6 34:4 23:8 IbmS 28:6 38:2 26:2 4.2 Performance Measures.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"basically complete, yet some improvements and extensions are still under way.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"However, there are phrases which express the same meanings even though they do not share the same keyword.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Our work is closest to that of Yarowsky and Ngai (2001), but differs in two important ways.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"See Table 2 for the tag set size of other languages.","label":0},{"source_text":"There is no global pruning.","target_text":"SSER: subjective sentence error rate: For a more detailed analysis, the translations are judged by a human test person.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"i..f,..","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We propose an unsupervised method to discover paraphrases from a large untagged corpus, without requiring any seed phrase or other cue.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"There is a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099 of ten commentaries, for which the range of information (except for syntax) has been completed; the remaining data has been annotated to different degrees, as explained below.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We adopted the MUC6 guidelines for evaluating coreference relationships based on transitivity in anaphoric chains.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"For all other recursive NPs, we add a common annotation to the POS tag of the head (recursiveNPHead).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We use N(u) to denote the neighborhood of vertex u, and fixed n = 5 in our experiments.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The German finite verbs 'bin' (second example) and 'konnten' (third example) are too far away from the personal pronouns 'ich' and 'Sie' (6 respectively 5 source sentence positions).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"needs to be in initCaps to be considered for this feature.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"and f,.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":".","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We weight edges using a unigram language model estimated with Good- Turing smoothing.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"92 76.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"However, when grammatical relations like subject and object are evaluated, parsing performance drops considerably (Green et al., 2009).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The choice between different actions is in general nondeterministic, and the parser relies on a memorybased classifier, trained on treebank data, to predict the next action based on features of the current parser configuration.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"It is also worth pointing out a connection with Daum\u00b4e\u2019s (2007) work that splits each feature into domain-specific and general copies.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Lossy compressed models RandLM (Talbot and Osborne, 2007) and Sheffield (Guthrie and Hepple, 2010) offer better memory consumption at the expense of CPU and accuracy.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"An important subproblem of language model storage is therefore sparse mapping: storing values for sparse keys using little memory then retrieving values given keys using little time.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Figure 2 shows a screenshot (which is of somewhat limited value, though, as color plays a major role in signalling the different statuses of the information).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"36.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"-1 means that an NP should be ruled out as a possible antecedent, and 0 means that the knowledge source remains neutral (i.e., it has no reason to believe that they cannot be coreferent).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Most similar is scrolling queries, wherein left-to-right queries that add one word at a time are optimized.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Text generation, or at least the two phases of text planning and sentence planning, is a process driven partly by well-motivated choices (e.g., use this lexeme X rather than that more colloquial near-synonym Y ) and partly by con tation like that of PCC can be exploited to look for correlations in particular between syntactic structure, choice of referring expressions, and sentence-internal information structure.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Our results suggest that current parsing models would benefit from better annotation consistency and enriched annotation in certain syntactic configurations.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"There is a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099 of ten commentaries, for which the range of information (except for syntax) has been completed; the remaining data has been annotated to different degrees, as explained below.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Mikheev et al.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"First we present the non-parametric version of parser switching, similarity switching: The intuition for this technique is that we can measure a similarity between parses by counting the constituents they have in common.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"As such, global information from the whole context of a document is important to more accurately recognize named entities.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Finally, a DempsterShafer probabilistic model evaluates the evidence provided by the knowledge sources for all candidate antecedents and makes the final resolution decision.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The most popular approach to dealing with seg\u00c2\u00ad mentation ambiguities is the maximum matching method, possibly augmented with further heuristics.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"To this end, we construct a bilingual graph over word types to establish a connection between the two languages (\u00a73), and then use graph label propagation to project syntactic information from English to the foreign language (\u00a74).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"25 16.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"All four of the techniques studied result in parsing systems that perform better than any previously reported.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The percentage scores on the axis labels represent the amount of variation in the data explained by the dimension in question.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"3.2 The DempsterShafer Decision Model.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"31 75.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"On each language we investigate the contribution of each component of our model.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"In general, the lemma of the previous section does not ensure that all the productions in the combined parse are found in the grammars of the member parsers.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"to explore how well we can induce POS tags using only the one-tag-per-word constraint.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The performance was 80.99% recall and 61.83% precision.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The machine learning community has been in a similar situation and has studied the combination of multiple classifiers (Wolpert, 1992; Heath et al., 1996).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The approach gains leverage from natural redundancy in the data: for many named-entity instances both the spelling of the name and the context in which it appears are sufficient to determine its type.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Tsarfaty and Sima\u2019an (2007) have reported state-of-the-art results on Hebrew unlexicalized parsing (74.41%) albeit assuming oracle morphological segmentation.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"No: Predecessor coverage set Successor coverage set 1 (f1; ;mg n flg ; l0) !","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Thus we opted not to take the step of creating more precise written annotation guidelines (as (Carlson, Marcu 2001) did for English), which would then allow for measuring inter-annotator agreement.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"As each global feature group is added to the list of features, we see improvements to both MUC6 and MUC6 MUC7 Baseline 90.75% 85.22% + ICOC 91.50% 86.24% + CSPP 92.89% 86.96% + ACRO 93.04% 86.99% + SOIC 93.25% 87.22% + UNIQ 93.27% 87.24% Table 3: F-measure after successive addition of each global feature group Table 5: Comparison of results for MUC6 Systems MUC6 MUC7 No.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Any opinions, findings, conclusions, or recommendations expressed in this paper are those of the authors, and do not necessarily reflect the views of the funding organizations.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The problem is to store these two values for a large and sparse set of n-grams in a way that makes queries efficient.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The developers suggested some changes, such as building the model from scratch with IRSTLM, but these did not resolve the problem.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Future work along these lines will incorporate other layers of annotation, in particular the syntax information.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The following rule corresponds to the above derivation, where 71, , 7k are derived from the auxiliary trees , , fik, respectively. for all addresses n in some elementary tree at which 7' can be adjoined.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thus, for example, one successor process will be have M to be in the existential state qa with the indices encoding xi , , xn, in the first 2n i tapes.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Table 9 shows that MADA produces a high quality segmentation, and that the effect of cascading segmentation errors on parsing is only 1.92% F1.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Unigram lookup is dense so we use an array of probability and backoff values.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"These methods demonstrated the benefits of incorporating linguistic features using a log-linear parameterization, but requires elaborate machinery for training.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In the next section we present an alternative approach that builds two classifiers while attempting to satisfy the above constraints as much as possible.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"This paper does not necessarily reflect the position of the U.S. Government.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In our forth model GTnph we add the definiteness status of constituents following Tsarfaty and Sima\u2019an (2007).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Hence, s < k. But by addition of the votes on the two parses, s > 2N-11> k, a contradiction.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The first issue relates to the completeness of the base lexicon.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Typical data structures are generalized Bloom filters that guarantee a customizable probability of returning the correct answer.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The system described in this paper is similar to the MENE system of (Borthwick, 1999).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Memory usage is likely much lower than ours. fThe original paper (Germann et al., 2009) provided only 2s of query timing and compared with SRI when it exceeded available RAM.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Finally, other packages implement language model estimation while we are currently dependent on them to generate an ARPA file.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Depending on the threshold t0, the search algorithm may miss the globally optimal path which typically results in additional translation errors.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"On average, 6 reference translations per automatic translation are available.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Contextual role knowledge provides evidence as to whether a candidate is a plausible antecedent for an anaphor.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Thus in a two-hanzi word like lflli?J zhong1guo2 (middle country) 'China' there are two syllables, and at the same time two morphemes.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"For monolingual treebank data we relied on the CoNLL-X and CoNLL-2007 shared tasks on dependency parsing (Buchholz and Marsi, 2006; Nivre et al., 2007).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Simply assigning to each word its most frequent associated tag in a corpus achieves 94.6% accuracy on the WSJ portion of the Penn Treebank.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The approach gains leverage from natural redundancy in the data: for many named-entity instances both the spelling of the name and the context in which it appears are sufficient to determine its type.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We use double-circles to indicate the space-delimited token boundaries.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"99 94.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We extend the Stanford parser to accept pre-generated lattices, where each word is represented as a finite state automaton.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The cost estimate, cost(i\u00c2\u00a5JJ1l.fn is computed in the obvious way by summing the negative log probabilities of i\u00c2\u00a5JJ1l.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"As a result, CFG's can not provide the structural descriptions in which there are nested dependencies between symbols labelling a path.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Evaluation results for links","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"5 60.6 Table 3: Multilingual Results: We report token-level one-to-one and many-to-one accuracy on a variety of languages under several experimental settings (Section 5).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"For English POS tagging, BergKirkpatrick et al. (2010) found that this direct gradient method performed better (>7% absolute accuracy) than using a feature-enhanced modification of the Expectation-Maximization (EM) algorithm (Dempster et al., 1977).8 Moreover, this route of optimization outperformed a vanilla HMM trained with EM by 12%.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Previous work has attempted to incorporate such constraints into token-level models via heavy-handed modifications to inference procedure and objective function (e.g., posterior regularization and ILP decoding) (Grac\u00c2\u00b8a et al., 2009; Ravi and Knight, 2009).","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Still, from a theoretical point of view, projective parsing of non-projective structures has the drawback that it rules out perfect accuracy even as an asymptotic goal.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Table 5: Effect of the beam threshold on the number of search errors (147 sentences).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Figure 2: An ATB sample from the human evaluation.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The algorithm builds two classifiers iteratively: each iteration involves minimization of a continuously differential function which bounds the number of examples on which the two classifiers disagree.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In our experiment, we set the threshold of the TF\/ITF score empirically using a small development corpus; a finer adjustment of the threshold could reduce the number of such keywords.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The algorithm can be viewed as heuristically optimizing an objective function suggested by (Blum and Mitchell 98); empirically it is shown to be quite successful in optimizing this criterion.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"See Section 5.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE without Proteus, however, did not do very well and only achieved an F measure of 84.22% (Borthwick, 1999).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Second, rather than relying on a division of the corpus into manually-assigned portions, we use features intended to capture the usefulness of each phrase pair.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Pseudo-labels are formed by taking seed labels on the labeled examples, and the output of the fixed classifier on the unlabeled examples.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Case and Zone of and : Similarly, if (or ) is initCaps, a feature (initCaps, zone) (or (initCaps, zone) ) is set to 1, etc. Token Information: This group consists of 10 features based on the string , as listed in Table 1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For a given \"word\" in the automatic segmentation, if at least k of the hu\u00c2\u00ad man judges agree that this is a word, then that word is considered to be correct.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Formally, let el (62) be the number of classification errors of the first (second) learner on the training data, and let Eco be the number of unlabeled examples on which the two classifiers disagree.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"However, it is robust, efficient, and easy to implement.4 To perform the maximization in (7), we used the popular L-BFGS algorithm (Liu and Nocedal, 1989), which requires gradient information.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"(2009).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(6), with W+ > W_.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Given a PCFG grammar G and a lattice L with nodes n1 ... nk, we construct the weighted grammar GL as follows: for every arc (lexeme) l E L from node ni to node nj, we add to GL the rule [l --+ tni, tni+1, ... , tnj_1] with a probability of 1 (this indicates the lexeme l spans from node ni to node nj).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"When extracting the vector t, used to compute the constraint feature from the graph, we tried three threshold values for r (see Eq.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"It is difficult when IN and OUT are dissimilar, as they are in the cases we study.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"2 We now introduce a new algorithm for learning from unlabeled examples, which we will call DLCoTrain (DL stands for decision list, the term Cotrain is taken from (Blum and Mitchell 98)).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The final block in table 2 shows models trained on feature subsets and on the SVM feature described in 3.4.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The translation direction is from German to English.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Nicola Bertoldi and Marcello Federico assisted with IRSTLM.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For corporate suffixes, a list of tokens cslist that occur frequently as the last token of an organization name is collected from the training data.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Sentence pairs are the natural instances for SMT, but sentences often contain a mix of domain-specific and general language.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Of course, since the number of attested (phonemic) Mandarin syllables (roughly 1400, including tonal distinctions) is far smaller than the number of morphemes, it follows that a given syllable could in principle be written with any of several different hanzi, depending upon which morpheme is intended: the syllable zhongl could be lfl 'middle,''clock,''end,' or ,'loyal.'","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"To be short, we omit the target words e; e0 in the formulation of the search hypotheses.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This offers the well-known advantages for inter- changability, but it raises the question of how to query the corpus across levels of annotation.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Keys to the table are hashed, using for example Austin Appleby\u2019s MurmurHash2, to integers evenly distributed over a large range.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"BABAR uses unsupervised learning to acquire this knowledge from plain text without the need for annotated training data.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This is a rather important source of errors in name identifi\u00c2\u00ad cation, and it is not really possible to objectively evaluate a name recognition system without considering the main lexicon with which it is used.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"(Kehler, 1997) also used a DempsterShafer model to merge evidence from different sources for template-level coreference.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Section 3 discusses the applications that have been completed with PCC, or are under way, or are planned for the future.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Table 1: Syntactic Seeding Heuristics BABAR\u00e2\u0080\u0099s reliable case resolution heuristics produced a substantial set of anaphor\/antecedent resolutions that will be the training data used to learn contextual role knowledge.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The results in Table 1 show PROBING is 81% faster than TRIE, which is in turn 31% faster than the fastest baseline.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"One is the accuracy within a set of phrases which share the same keyword; the other is the accuracy of links.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The tee pumping lemma states that if there is tree, t = 22 t2t3, generated by a CFG G, whose height is more than a predetermined bound k, then all trees of the form ti tP3 for each i > 0 will also generated by G (as shown in Figure 9b).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Context from the whole document can be important in classifying a named entity.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"An alternate approximation to (8) would be to let w,\\(s, t) directly approximate p\u02c6I(s, t).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"An inverted alignment is defined as follows: inverted alignment: i ! j = bi: Target positions i are mapped to source positions bi.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Borth- wick (1999) successfully made use of other hand- coded systems as input for his MENE system, and achieved excellent results.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Gather phrases using keywords Next, we select a keyword for each phrase \u00e2\u0080\u0093 the top-ranked word based on the TF\/IDF metric.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The relevance of the distinction between, say, phonological words and, say, dictionary words is shown by an example like rpftl_A :;!:Hfllil zhong1hua2 ren2min2 gong4he2-guo2 (China people republic) 'People's Republic of China.'","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We do not experiment with models larger than physical memory in this paper because TPT is unreleased, factors such as disk speed are hard to replicate, and in such situations we recommend switching to a more compact representation, such as RandLM.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Once the lexicon has been drawn, the model proceeds similarly to the standard token-level HMM: Emission parameters \u00ce\u00b8 are generated conditioned on tag assignments T . We also draw transition parameters \u00cf\u0086.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"2.5 Connectives with scopes.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"1 74.5 56.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In order to train this error-correction model, he divided his training corpus into 5 portions of 20% each.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"3.2 Stochastic rhetorical analysis.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"ments contained 322 anaphoric links.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The label for a test example with features x is then defined as In this paper we define h(x, y) as the following function of counts seen in training data: Count(x,y) is the number of times feature x is seen with label y in training data, Count(x) = EyEy Count(x, y). a is a smoothing parameter, and k is the number of possible labels.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This highly effective approach is not directly applicable to the multinomial models used for core SMT components, which have no natural method for combining split features, so we rely on an instance-weighting approach (Jiang and Zhai, 2007) to downweight domain-specific examples in OUT.","label":0},{"source_text":"There is no global pruning.","target_text":"The alignment model uses two kinds of parameters: alignment probabilities p(aj jaj\u00f4\u0080\u0080\u00801; I; J), where the probability of alignment aj for position j depends on the previous alignment position aj\u00f4\u0080\u0080\u00801 (Ney et al., 2000) and lexicon probabilities p(fj jeaj ).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Because many foreign word types are not aligned to an English word (see Table 3), and we do not run label propagation on the foreign side, we expect the projected information to have less coverage.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Our experiments consistently demonstrate that this model architecture yields substantial performance gains over more complex tagging counterparts.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"was done by the participants.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In such cases we assign all of the estimated probability mass to the form with the most likely pronunciation (determined by inspection), and assign a very small probability (a very high cost, arbitrarily chosen to be 40) to all other variants.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For example, in MUC6, there are four zones (TXT, HL, DATELINE, DD).","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"However, this argument is only plausible if the formal framework allows non-projective dependency structures, i.e. structures where a head and its dependents may correspond to a discontinuous constituent.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"A simple extension will be used to handle this problem.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This essentially means that their model does not result in a true joint inference, as executions for different tasks involve tuning a parameter separately.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"It may seem surprising to some readers that the interhuman agreement scores reported here are so low.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Borth 2MUC data can be obtained from the Linguistic Data Consortium: http:\/\/www.ldc.upenn.edu 3Training data for IdentiFinder is actually given in words (i.e., 650K & 790K words), rather than tokens Table 6: Comparison of results for MUC7 wick (1999) reported using dictionaries of person first names, corporate names and suffixes, colleges and universities, dates and times, state abbreviations, and world regions.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"com \u00c2\u00a7Cambridge, UK Email: nc201@eng.cam.ac.uk \u00c2\u00a9 1996 Association for Computational Linguistics (a) B ) ( , : & ; ? ' H o w d o y o u s a y o c t o p u s i n J a p a n e s e ? ' (b) P l a u s i b l e S e g m e n t a t i o n I B X I I 1 : & I 0 0 r i 4 w e n 2 z h a n g l y u 2 z e n 3 m e 0 s h u o l ' J a p a n e s e ' ' o c t o p u s ' ' h o w ' ' s a y ' (c) Figure 1 I m p l a u s i b l e S e g m e n t a t i o n [\u00c2\u00a7] lxI 1:&I ri4 wen2 zhangl yu2zen3 me0 shuol 'Japan' 'essay' 'fish' 'how' 'say' A Chinese sentence in (a) illustrating the lack of word boundaries.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In this situation, BABAR takes the conservative approach and declines to make a resolution.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The choice of the genre commentary resulted from the fact that an investigation of rhetorical structure, its interaction with other aspects of discourse structure, and the prospects for its automatic derivation are the key motivations for building up the corpus.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The CoBoost algorithm just described is for the case where there are two labels: for the named entity task there are three labels, and in general it will be useful to generalize the CoBoost algorithm to the multiclass case.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We adopted the MUC6 guidelines for evaluating coreference relationships based on transitivity in anaphoric chains.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"For each language and setting, we report one-to-one (11) and many- to-one (m-1) accuracies.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR employs information extraction techniques to represent and learn role relationships.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"For developing these mechanisms, the possibility to feed in hand-annotated information is very useful.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"It is natural that the larger the data in the domain, the more keywords are found.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"MENE (Maximum Entropy Named Entity) (Borth- wick, 1999) was combined with Proteus (a hand- coded system), and came in fourth among all MUC 7 participants.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Tokens were converted to vocabulary identifiers in advance and state was carried from each query to the next.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In that work, mutual information was used to decide whether to group adjacent hanzi into two-hanzi words.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Te rro ris m Na tur al Dis ast ers mu rde r of < NP > kill ed <p atie nt > <a ge nt > da ma ged wa s inj ure d in < NP > <a ge nt > rep ort ed <a ge nt > add ed <a ge nt > occ urr ed cau se of < NP > <a ge nt > stat ed <a ge nt > add ed <a ge nt > wr eak ed <a ge nt > cro sse d per pet rat ed <p atie nt > con de mn ed <p atie nt > dri ver of < NP > <a ge nt > car ryi ng Figure 1: Caseframe Network Examples Figure 1 shows examples of caseframes that co-occur in resolutions, both in the terrorism and natural disaster domains.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"There is no relation between these two interpretations other then the fact that their surface forms coincide, and we argue that the only reason to prefer one analysis over the other is compositional.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We thank United Informatics for providing us with our corpus of Chinese text, and BDC for the 'Behavior ChineseEnglish Electronic Dictionary.'","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"kann 7.nicht 8.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"One knowledge source, called WordSemCFSem, is analogous to CFLex: it checks whether the anaphor and candidate antecedent are substitutable for one another, but based on their semantic classes instead of the words themselves.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Our second contribution is to apply instance weighting at the level of phrase pairs.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We also thank ChaoHuang Chang, reviewers for the 1994 ACL conference, and four anonymous reviewers for Computational Linguistics for useful comments.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"In the label propagation stage, we propagate the automatic English tags to the aligned Italian trigram types, followed by further propagation solely among the Italian vertices. the Italian vertices are connected to an automatically labeled English vertex.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The last four columns in Table 3 show the distribution of nonprojective arcs with respect to the number of lifts required.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Links can be of two different kinds: anaphoric or bridging (definite noun phrases picking up an antecedent via world-knowledge).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The dictionary sizes reported in the literature range from 17,000 to 125,000 entries, and it seems reasonable to assume that the coverage of the base dictionary constitutes a major factor in the performance of the various approaches, possibly more important than the particular set of methods used in the segmentation.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"(Thus the domain of the dev and test corpora matches IN.)","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Central to our approach (see Algorithm 1) is a bilingual similarity graph built from a sentence-aligned parallel corpus.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The negative logarithm of t0 is reported.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"An ATM has two types of states, existential and universal.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Figure 2: An ATB sample from the human evaluation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"52 15.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Dropping the conditioning on 0 for brevity, and letting \u00afc\u03bb(s, t) = c\u03bb(s, t) + yu(s|t), and \u00afc\u03bb(t) = 4Note that the probabilities in (7) need only be evaluated over the support of \u02dcp(s, t), which is quite small when this distribution is derived from a dev set.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Finally, note that while most feature concepts are lexicalized, others, such as the suffix concept, are not.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Furthermore we expect the label distributions on the foreign to be fairly noisy, because the graph constraints have not been taken into account yet.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"This means that the PCC cannot grow particularly quickly.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"67 95.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Label propagation can then be used to transfer the labels to the peripheral foreign vertices (i.e. the ones adjacent to the English vertices) first, and then among all of the foreign vertices (\u00a74).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Feature selection is implemented using a feature cutoff: features seen less than a small count during training will not be used.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"As described in Sproat (1995), the Chinese segmenter presented here fits directly into the context of a broader finite-state model of text analysis for speech synthesis.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Training examples are generated automatically by identifying noun phrases that can be easily resolved with their antecedents using lexical and syntactic heuristics.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Several such particles may be prefixed onto a single stem, in which case the affixation is subject to strict linear precedence constraints.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"3 Throughout this paper we shall give Chinese examples in traditional orthography, followed.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"If either case is true, then CFLex reports that the anaphor and candidate might be coreferent.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Once we figure out the important word (e.g. keyword), we believe we can capture the meaning of the phrase by the keyword.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"This solution also obviates the need to perform word sense disambiguation.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"64 76.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The range of the score is between 0 and 1 (higher is better).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"One class comprises words derived by productive morphologi\u00c2\u00ad cal processes, such as plural noun formation using the suffix ir, menD.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We evaluate our model on seven languages exhibiting substantial syntactic variation.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In general, different modalities (\u00e2\u0080\u009cplanned to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cagreed to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cbought\u00e2\u0080\u009d) were considered to express the same relationship within an extraction setting.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The Input The set of analyses for a token is thus represented as a lattice in which every arc corresponds to a specific lexeme l, as shown in Figure 1.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"level.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"In order to evaluate and advance this approach, it helps to feed into the knowledge base data that is already enriched with some of the desired information \u00e2\u0080\u0094 as in PCC.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Third, we develop a human interpretable grammar that is competitive with a latent variable PCFG.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Our system does not currently make use of titles, but it would be straightforward to do so within the finite-state framework that we propose.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"One is the accuracy within a set of phrases which share the same keyword; the other is the accuracy of links.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"That is, we can use the discourse parser on PCC texts, emulating for instance a \u00e2\u0080\u009cco-reference oracle\u00e2\u0080\u009d that adds the information from our co-reference annotations.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Our approach uses word-to-word dependencies between source and target words.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Interestingly, Chang et al. report 80.67% recall and 91.87% precision on an 11,000 word corpus: seemingly, our system finds as many names as their system, but with four times as many false hits.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Nonetheless, the prospect of a network of annotated discourse resources seems particularly promising if not only a single annotation layer is used but a whole variety of them, so that a systematic search for correlations between them becomes possible, which in turn can lead to more explanatory models of discourse structure.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In this paper, we show that the maximum entropy framework is able to make use of global information directly, and achieves performance that is comparable to the best previous machine learning-based NERs on MUC6 and MUC7 test data.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"It did 402 queries\/ms using 1.80 GB. cMemory use increased during scoring due to batch processing (MIT) or caching (Rand).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Our test machine has two Intel Xeon E5410 processors totaling eight cores, 32 GB RAM, and four Seagate Barracuda disks in software RAID 0 running Linux 2.6.18.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"A corpus of German newspaper commentaries has been assembled at Potsdam University, and annotated with different linguistic information, to different degrees.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In our grammar, features are realized as annotations to basic category labels.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Here, \u00e2\u0080\u009cEG\u00e2\u0080\u009d represents \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Table 2 shows results for both settings and all methods described in sections 2 and 3.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Certainly these linguistic factors increase the difficulty of syntactic disambiguation.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We maintain a separate array for each length n containing all n-gram entries sorted in suffix order.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Thus, the length of any string in L is a linear combination of the length of strings in some fixed finite subset of L, and thus L is said to have the constant growth property.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"By making use of global context, it has achieved excellent results on both MUC6 and MUC7 official test data.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Covering the first uncovered position in the source sentence, we use the language model probability p(ej$; $).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Proper names are assumed to be coreferent if they match exactly, or if they closely match based on a few heuristics.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"0n0'i'i0'2&quot;bin242bn I n = 711 + n2 } On the other hand, no linguistic use is made of this general form of composition and Steedman (personal communication) and Steedman (1986) argues that a more limited definition of composition is more natural.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.2.2 The Caseframe Network The first type of contextual role knowledge that BABAR learns is the Caseframe Network (CFNet), which identifies caseframes that co-occur in anaphor\/antecedent resolutions.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"input: source string f1:::fj :::fJ initialization for each cardinality c = 1; 2; ; J do for each pair (C; j), where j 2 C and jCj = c do for each target word e 2 E Qe0 (e; C; j) = p(fj je) max \u00c3\u0086;e00 j02Cnfjg fp(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j0)g words fj in the input string of length J. For the final translation each source position is considered exactly once.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"More details on the parsing algorithm can be found in Nivre (2003).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Dagan and Itai (Dagan and Itai, 1990) experimented with co-occurrence statistics that are similar to our lexical caseframe expectations.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"While the paper mentioned a sorted variant, code was never released.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We refer to this process as Reliable Case Resolution because it involves finding cases of anaphora that can be easily resolved with their antecedents.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We attain these results using several optimizations: hashing, custom lookup tables, bit-level packing, and state for left-to-right query patterns.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"We propose maximizing , where is the sequence of named- entity tags assigned to the words in the sentence , and is the information that can be extracted from the whole document containing . Our system is built on a maximum entropy classifier.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Morphological disambiguators that consider a token in context (an utterance) and propose the most likely morphological analysis of an utterance (including segmentation) were presented by Bar-Haim et al. (2005), Adler and Elhadad (2006), Shacham and Wintner (2007), and achieved good results (the best segmentation result so far is around 98%).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The P (W |T , \u00cf\u0088) term in the lexicon component now decomposes as: n P (W |T , \u00cf\u0088) = n P (Wi|Ti, \u00cf\u0088) i=1 n \u00ef\u00a3\u00ab \u00ef\u00a3\u00b6 tions are not modeled by the standard HMM, which = n \u00ef\u00a3\u00ad n P (v|\u00cf\u0088Ti f )\u00ef\u00a3\u00b8 instead can model token-level frequency.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Tag set As is standard, for all experiments, we set the number of latent model tag states to the size of the annotated tag set.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The learned patterns are then normalized and applied to the corpus.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The productions of HG's are very similar to those of CFG's except that the operation used must be made explicit.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(See Sproat and Shih 1995.)","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"\u00c2\u00a3 : _ADV: 5.88 If:!","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In TAG's the elementary tree and addresses where adjunction takes place are used to instantiate the operation.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"However, the data sparsity induced by vocalization makes it difficult to train statistical models on corpora of the size of the ATB, so vocalizing and then parsing may well not help performance.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Although this is not a precise criterion, most cases we evaluated were relatively clear-cut.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"On the first of these-the B set-our system had 64% recall and 86% precision; on the second-the C set-it had 33% recall and 19% precision.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"This paper is based on work supported in part by DARPA through IBM.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If they are found in a list, then a feature for that list will be set to 1.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The derivation trees of a MCTAG are similar to those of a TAG.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"4.1 The Task and the Corpus.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The tokens w are generated by token-level tags t from an HMM parameterized by the lexicon structure.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The \u201cNo LP\u201d model does not outperform direct projection for German and Greek, but performs better for six out of eight languages.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Feature-based HMM Model (Berg- Kirkpatrick et al., 2010): The KM model uses a variety of orthographic features and employs the EM or LBFGS optimization algorithm; Posterior regulariation model (Grac\u00c2\u00b8a et al., 2009): The G10 model uses the posterior regular- ization approach to ensure tag sparsity constraint.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"2 56.2 32.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"But diacritics are not present in unvocalized text, which is the standard form of, e.g., news media documents.3 VBD she added VP PUNC S VP VBP NP ...","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We will briefly discuss this point in Section 3.1.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"It is also true of the adaptation of the Collins parser for Czech (Collins et al., 1999) and the finite-state dependency parser for Turkish by Oflazer (2003).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Notice that the CC-domain is a special case.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We have presented two general approaches to studying parser combination: parser switching and parse hybridization.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We extend Subramanya et al.\u2019s intuitions to our bilingual setup.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Yet we note that the better grammars without pruning outperform the poorer grammars using this technique, indicating that the syntactic context aids, to some extent, the disambiguation of unknown tokens.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We have estimated the performance of IdentiFinder ' 99 at 200K words of training data from the graphs.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The maximum precision row is the upper bound on accuracy if we could pick exactly the correct constituents from among the constituents suggested by the three parsers.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Both of the switching techniques, as well as the parametric hybridization technique were also shown to be robust when a poor parser was introduced into the experiments.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"of Arabic.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Formally, the DempsterShafer theory defines a probability density function m(S), where S is a set of hypotheses.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Also, we don\u00e2\u0080\u0099t know how many such paraphrase sets are necessary to cover even some everyday things or events.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"To differentiate between the coordinating and discourse separator functions of conjunctions (Table 3), we mark each CC with the label of its right sister (splitCC).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"gaolxing4 'happy' => F.i'JF.i'J Jl!","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The 1st block contains the simple baselines from section 2.1.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Table 3 shows BABAR\u00e2\u0080\u0099s performance when the four contextual role knowledge sources are added.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Now assume we have n pairs (xi,, x2,i) drawn from X1 X X2, where the first m pairs have labels whereas for i = m+ 1...n the pairs are unlabeled.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"However, the point of RandLM is to scale to even larger data, compensating for this loss in quality.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Second, comparisons of different methods are not meaningful unless one can eval\u00c2\u00ad uate them on the same corpus.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Several methods exist to handle collisions; we use linear probing because it has less memory overhead when entries are small.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"This number must be less than or equal to n \u00f4\u0080\u0080\u0080 1.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We have presented two general approaches to studying parser combination: parser switching and parse hybridization.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"63 95.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We have developed a general approach for combining parsers when preserving the entire structure of a parse tree is important.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"(f1; ;mg n fl1; l2g ; l) 4 (f1; ;m \u00f4\u0080\u0080\u0080 1g n fl1; l2; l3g ; l0) !","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Here, an NE instance pair is any pair of NEs separated by at most 4 syntactic chunks; for example, \u00e2\u0080\u009cIBM plans to acquire Lotus\u00e2\u0080\u009d.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Alternatively, h can be thought of as defining a decision list of rules x y ranked by their &quot;strength&quot; h(x, y).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Test set OOV rate is computed using the following splits: ATB (Chiang et al., 2006); CTB6 (Huang and Harper, 2009); Negra (Dubey and Keller, 2003); English, sections 221 (train) and section 23 (test).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We used section 23 as the development set for our combining techniques, and section 22 only for final testing.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Note that Zt is a normalization constant that ensures the distribution Dt+i sums to 1; it is a function of the weak hypothesis ht and the weight for that hypothesis at chosen at the tth round.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In these examples, the names identified by the two systems (if any) are underlined; the sentence with the correct segmentation is boxed.19 The differences in performance between the two systems relate directly to three issues, which can be seen as differences in the tuning of the models, rather than repre\u00c2\u00ad senting differences in the capabilities of the model per se.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"An example of a fairly low-level relation is the affix relation, which holds between a stem morpheme and an affix morpheme, such as f1 -menD (PL).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"TRIE uses less memory and has better locality.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Among the IS-units, the referring expressions are marked as such and will in the second phase receive a label for cognitive status (active, accessible- text, accessible-situation, inferrable, inactive).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"(2009) on Portuguese (Grac\u00c2\u00b8a et al.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We would like to relax somewhat the constraint on the path complexity of formalisms in LCFRS.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"On a set of 11 sentence fragments-the A set-where they reported 100% recall and precision for name identification, we had 73% recall and 80% precision.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"One possibility is to use n-grams based on mutual information.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"One hybridization strategy is to let the parsers vote on constituents' membership in the hypothesized set.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Across all languages, +PRIOR consistently outperforms 1TW, reducing error on average by 9.1% and 5.9% on best and median settings respectively.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"This means that the PCC cannot grow particularly quickly.","label":0},{"source_text":"There is no global pruning.","target_text":"German city names are replaced by category markers.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"While the paper mentioned a sorted variant, code was never released.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"7).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We have developed a general approach for combining parsers when preserving the entire structure of a parse tree is important.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"As we have said, parse quality decreases with sentence length.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Thus, provided at least this amount of IN data is available\u2014as it is in our setting\u2014adapting these weights is straightforward.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Clearly this is not the only way to estimate word-frequencies, however, and one could consider applying other methods: in partic\u00c2\u00ad ular since the problem is similar to the problem of assigning part-of-speech tags to an untagged corpus given a lexicon and some initial estimate of the a priori probabilities for the tags, one might consider a more sophisticated approach such as that described in Kupiec (1992); one could also use methods that depend on a small hand-tagged seed corpus, as suggested by one reviewer.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Two lists, Corporate-Suffix-List (for corporate suffixes) and Person-Prefix-List (for person prefixes), are collected from the training data.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For example, hanzi containing the INSECT radical !R tend to denote insects and other crawling animals; examples include tr wal 'frog,' feng1 'wasp,' and !Itt she2 'snake.'","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"(Webber et al., 2003)).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Timing is based on plentiful memory.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Instead, we condition on the type-level tag assignments T . Specifically, let St = {i|Ti = t} denote the indices of theword types which have been assigned tag t accord ing to the tag assignments T . Then \u00ce\u00b8t is drawn from DIRICHLET(\u00ce\u00b1, St), a symmetric Dirichlet which only places mass on word types indicated by St. This ensures that each word will only be assigned a single tag at inference time (see Section 4).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.1 Part-of-speech tags.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Jiang and Zhai (2007) suggest the following derivation, making use of the true OUT distribution po(s, t): where each fi(s, t) is a feature intended to charac- !0\u02c6 = argmax pf(s, t) log p\u03b8(s|t) (8) terize the usefulness of (s, t), weighted by Ai. \u03b8 s,t pf(s, t)po(s, t) log p\u03b8(s|t) The mixing parameters and feature weights (col- != argmax po (s, t) lectively 0) are optimized simultaneously using dev- \u03b8 s,t pf(s, t)co(s, t) log p\u03b8(s|t), set maximum likelihood as before: !\ufffdargmax po (s, t) ! \u03b8 s,t \ufffd\u02c6 = argmax \u02dcp(s, t) log p(s|t; 0).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Analysis of the data revealed that the contextual role knowledge is especially helpful for resolving pronouns because, in general, they are semantically weaker than definite NPs.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"buy - acquire (5) buy - agree (2) buy - purchase (5) buy - acquisition (7) buy - pay (2)* buy - buyout (3) buy - bid (2) acquire - purchase (2) acquire - acquisition (2) acquire - pay (2)* purchase - acquisition (4) purchase - stake (2)* acquisition - stake (2)* unit - subsidiary (2) unit - parent (5) It is clear that these links form two clusters which are mostly correct.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(1992).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We asked six native speakers-three from Taiwan (TlT3), and three from the Mainland (M1M3)-to segment the corpus.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The choice of the particular newspaper was motivated by the fact that the language used in a regional daily is somewhat simpler than that of papers read nationwide.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"This approach is compared to another reordering scheme presented in (Berger et al., 1996).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Recent results (e.g., (Yarowsky 95; Brill 95; Blum and Mitchell 98)) have suggested that unlabeled data can be used quite profitably in reducing the need for supervision.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(1992).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We also collapse unary chains withidentical basic categories like NP \u00e2\u0086\u0092 NP.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A position is presented by the word at that position.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"First, from a large corpus, we extract all the NE instance pairs.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"So far the graph has been completely unlabeled.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For previously unseen hanzi in given names, Chang et al. assign a uniform small cost; but we know that some unseen hanzi are merely acci\u00c2\u00ad dentally missing, whereas others are missing for a reason-for example, because they have a bad connotation.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This might be because our features are more comprehensive than those used by Borthwick.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"For developing these mechanisms, the possibility to feed in hand-annotated information is very useful.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For example, if a token starts with a capital letter and ends with a period (such as Mr.), then the feature InitCapPeriod is set to 1, etc. First Word: This feature group contains only one feature firstword.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"2.4 Underspecified rhetorical structure.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For example, one of the ATB samples was the determiner -\"\" ; dhalik\u00e2\u0080\u009cthat.\u00e2\u0080\u009d The sample occurred in 1507 corpus po sitions, and we found that the annotations were consistent.","label":0},{"source_text":"There is no global pruning.","target_text":"The details are given in (Tillmann, 2000).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"IRSTLM 5.60.02 (Federico et al., 2008) is a sorted trie implementation designed for lower memory consumption.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In German, the verbgroup usually consists of a left and a right verbal brace, whereas in English the words of the verbgroup usually form a sequence of consecutive words.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"A corpus of German newspaper commentaries has been assembled at Potsdam University, and annotated with different linguistic information, to different degrees.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Further, we report current resident memory and peak virtual memory because these are the most applicable statistics provided by the kernel.","label":0},{"source_text":"There is no global pruning.","target_text":"To explicitly handle the word reordering between words in source and target language, we use the concept of the so-called inverted alignments as given in (Ney et al., 2000).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The knowledge base then can be tested for its relation-inference capabilities on the basis of full-blown co-reference information.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"7).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2010) reports the best unsupervised results for English.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The availability of comparable corpora is limited, which is a significant limitation on the approach.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"A secondary reference resolution classifier has information on the class assigned by the primary classifier.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"encodes the one tag per word constraint and is uni form over type-level tag assignments.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"To initialize the graph for label propagation we use a supervised English tagger to label the English side of the bitext.7 We then simply count the individual labels of the English tokens and normalize the counts to produce tag distributions over English word types.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"5 We choose these two metrics over the Variation Information measure due to the deficiencies discussed in Gao and Johnson (2008).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"KS Function Ge nde r filters candidate if gender doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"However their work did not consider other types of lexical expectations (e.g., PP arguments), semantic expectations, or context comparisons like our case- frame network.(Niyu et al., 1998) used unsupervised learning to ac quire gender, number, and animacy information from resolutions produced by a statistical pronoun resolver.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The final model tions.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We generate these caseframes automatically by running AutoSlog over the training corpus exhaustively so that it literally generates a pattern to extract every noun phrase in the corpus.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"As a result, Habash et al.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"This means that the PCC cannot grow particularly quickly.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The motivation for our more informal approach was the intuition that there are so many open problems in rhetorical analysis (and more so for German than for English; see below) that the main task is qualitative investigation, whereas rigorous quantitative analyses should be performed at a later stage.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Evaluation Metrics We report three metrics to evaluate tagging performance.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Besides size of training data, the use of dictionaries is another factor that might affect performance.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"We also see that the increase in the size of the label sets for Head and Head+Path is far below the theoretical upper bounds given in Table 1.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We model each parse as the decisions made to create it, and model those decisions as independent events.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Within the RST \u00e2\u0080\u009cuser community\u00e2\u0080\u009d there has also been discussion whether two levels of discourse structure should not be systematically distinguished (intentional versus informational).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We picked two domains, the CC-domain and the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain (PC-domain), for the evaluation, as the entire system output was too large to evaluate.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Assigning rhetorical relations thus poses questions that can often be answered only subjectively.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"This might be because our features are more comprehensive than those used by Borthwick.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In total 13,976 phrases are assigned to sets of phrases, and the accuracy on our evaluation data ranges from 65 to 99%, depending on the domain and the size of the sets.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"A contextual rule considers words surrounding the string in the sentence in which it appears (e.g., a rule that any proper name modified by an appositive whose head is president is a person).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Our evaluation includes both weighted and un- weighted lattices.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We show the results of three of the experiments we conducted to measure isolated constituent precision under various partitioning schemes.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The sentence-selection approach is crude in that it imposes a binary distinction between useful and non-useful parts of OUT.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"First, we aim to explicitly characterize examples from OUT as belonging to general language or not.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Same for . In the case where the next token is a hyphen, then is also used as a feature: (init- Caps, ) is set to 1.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The corpora for both settings are summarized in table 1.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"For example, out of 905 phrases in the CC- domain, 211 phrases contain keywords found in step 2.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We use Alternating Turing Machines (Chandra, Kozen, and Stockmeyer, 1981) to show that polynomial time recognition is possible for the languages discussed in Section 4.3.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The approach builds from an initial seed set for a category, and is quite similar to the decision list approach described in (Yarowsky 95).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"5 68.1 34.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"As is standard, we report the greedy one-to-one (Haghighi and Klein, 2006) and the many-to-one token-level accuracy obtained from mapping model states to gold POS tags.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"During the search process, a partial hypothesis is extended by choosing a source sentence position, which has not been aligned with a target sentence position yet.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Our annotators pointed out that very often they made almost random decisions as to what relation to choose, and where to locate the boundary of a span.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The larger sets are more accurate than the small sets.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Having explained the various layers of annotation in PCC, we now turn to the question what all this might be good for.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Figure 1 shows sample sentences from these domains, which are widely divergent.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"set was based on an earlier version of the Chang et a!.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"1 61.7 37.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The paper explains the design decisions taken in the annotations, and describes a number of applications using this corpus with its multi-layer annotation.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"2 56.2 32.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"When this feature type was included, CoBoost chose this default feature at an early iteration, thereby giving non-abstaining pseudo-labels for all examples, with eventual convergence to the two classifiers agreeing by assigning the same label to almost all examples.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The results in Table 2 were achieved on the development set.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":".","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Foreign names are usually transliterated using hanzi whose sequential pronunciation mimics the source language pronunciation of the name.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Borth 2MUC data can be obtained from the Linguistic Data Consortium: http:\/\/www.ldc.upenn.edu 3Training data for IdentiFinder is actually given in words (i.e., 650K & 790K words), rather than tokens Table 6: Comparison of results for MUC7 wick (1999) reported using dictionaries of person first names, corporate names and suffixes, colleges and universities, dates and times, state abbreviations, and world regions.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"72 77.","label":0},{"source_text":"Here we present two algorithms.","target_text":"To see this, note thai the first two terms in the above equation correspond to the function that AdaBoost attempts to minimize in the standard supervised setting (Equ.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"For other languages, we use the CoNLL-X multilingual dependency parsing shared task corpora (Buchholz and Marsi, 2006) which include gold POS tags (used for evaluation).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"D o m ai n # of ph ras es t o t a l p h r a s e s ac cu ra cy C C 7 o r m o r e 1 0 5 8 7 . 6 % 6 o r l e s s 1 0 6 6 7 . 0 % P C 7 o r m o r e 3 5 9 9 9 . 2 % 6 o r l e s s 2 5 5 6 5 . 1 % Table 1.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Step 2.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Log-linear combination (loglin) improves on this in all cases, and also beats the pure IN system.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"A token that is allCaps will also be initCaps.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For MUC6, for example, Corporate- Suffix-List is made up of ltd., associates, inc., co, corp, ltd, inc, committee, institute, commission, university, plc, airlines, co., corp. and Person-Prefix- List is made up of succeeding, mr., rep., mrs., secretary, sen., says, minister, dr., chairman, ms. . For a token that is in a consecutive sequence of init then a feature Corporate-Suffix is set to 1.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Figure 1 reveals that an event that \u00e2\u0080\u009cdamaged\u00e2\u0080\u009d objects may also cause injuries; a disaster that \u00e2\u0080\u009coccurred\u00e2\u0080\u009d may be investigated to find its \u00e2\u0080\u009ccause\u00e2\u0080\u009d; a disaster may \u00e2\u0080\u009cwreak\u00e2\u0080\u009d havoc as it \u00e2\u0080\u009ccrosses\u00e2\u0080\u009d geographic regions; and vehicles that have a \u00e2\u0080\u009cdriver\u00e2\u0080\u009d may also \u00e2\u0080\u009ccarry\u00e2\u0080\u009d items.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Automatic Paraphrase Discovery based on Context and Keywords between NE Pairs","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In an existential state an ATM behaves like a nondeterministic TM, accepting if one of the applicable moves leads to acceptance; in an universal state the ATM accepts if all the applicable moves lead to acceptance.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Furthermore, some of the arcs represent lexemes not present in the input tokens (e.g. h\/DT, fl\/POS), however these are parts of valid analyses of the token (cf. super-segmental morphology section 2).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The algorithm, called CoBoost, has the advantage of being more general than the decision-list learning alInput: (xi , yi), , (xim, ) ; x, E 2x, yi = +1 Initialize Di (i) = 1\/m.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Our test machine has two Intel Xeon E5410 processors totaling eight cores, 32 GB RAM, and four Seagate Barracuda disks in software RAID 0 running Linux 2.6.18.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"For each feature type f and tag t, a multinomial \u00cf\u0088tf is drawn from a symmetric Dirichlet distribution with concentration parameter \u00ce\u00b2.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The model starts by generating a tag assignment for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"8 We use head-finding rules specified by a native speaker.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"While our method also enforces a singe tag per word constraint, it leverages the transition distribution encoded in an HMM, thereby benefiting from a richer representation of context.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"html 4 www.wagsoft.com\/RSTTool assigning rhetorical relations is a process loaded with ambiguity and, possibly, subjectivity.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"A dynamic programming recursion similar to the one in Eq. 2 is evaluated.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Jiang and Zhai (2007) suggest the following derivation, making use of the true OUT distribution po(s, t): where each fi(s, t) is a feature intended to charac- !0\u02c6 = argmax pf(s, t) log p\u03b8(s|t) (8) terize the usefulness of (s, t), weighted by Ai. \u03b8 s,t pf(s, t)po(s, t) log p\u03b8(s|t) The mixing parameters and feature weights (col- != argmax po (s, t) lectively 0) are optimized simultaneously using dev- \u03b8 s,t pf(s, t)co(s, t) log p\u03b8(s|t), set maximum likelihood as before: !\ufffdargmax po (s, t) ! \u03b8 s,t \ufffd\u02c6 = argmax \u02dcp(s, t) log p(s|t; 0).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Top 5 Bot to m 5 Go ld NN P NN JJ CD NN S RB S PD T # \u00e2\u0080\u009d , 1T W CD W RB NN S VB N NN PR P$ W DT : MD . +P RI OR CD JJ NN S WP $ NN RR B- , $ \u00e2\u0080\u009d . +F EA TS JJ NN S CD NN P UH , PR P$ # . \u00e2\u0080\u009c Table 5: Type-level English POS Tag Ranking: We list the top 5 and bottom 5 POS tags in the lexicon and the predictions of our models under the best hyperparameter setting.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The question of what soft function to pick, and how to design' algorithms which optimize it, is an open question, but appears to be a promising way of looking at the problem.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For example, management succession systems must distinguish between a person who is fired and a person who is hired.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For each candidate antecedent, BABAR identifies the caseframe that would extract the candidate, pairs it with the anaphor\u00e2\u0080\u0099s caseframe, and consults the CF Network to see if this pair of caseframes has co-occurred in previous resolutions.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"splitIN captures the verb\/preposition idioms that are widespread in Arabic.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Terrorism systems must distinguish between people who perpetrate a crime and people who are victims of a crime.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"information structure.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If is not initCaps, then (not-initCaps, ) is set to 1.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"(we ignored the 419 trees in their development set.)","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We propose an unsupervised method to discover paraphrases from a large untagged corpus, without requiring any seed phrase or other cue.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"4.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If any of the tokens from to is in Person-Prefix- List, then another feature Person-Prefix is set to 1.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Like in the co-reference annotation, G\u00c2\u00a8otze\u00e2\u0080\u0099s proposal has been applied by two annotators to the core corpus but it has not been systematically evaluated yet.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For example, suppose one is building a ITS system for Mandarin Chinese.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We would like to thank Ryan McDonald for numerous discussions on this topic.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"set was based on an earlier version of the Chang et a!.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We have estimated the performance of IdentiFinder ' 99 at 200K words of training data from the graphs.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"3.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Before describing the unsupervised case we first describe the supervised version of the algorithm: Input to the learning algorithm: n labeled examples of the form (xi, y\u201e). y, is the label of the ith example (given that there are k possible labels, y, is a member of y = {1 ... 0). xi is a set of mi features {x,1, Xi2 .","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Unlike Germann et al. (2009), we chose a model size so that all benchmarks fit comfortably in main memory.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Final (F): The rest of the sentence is processed monotonically taking account of the already covered positions.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"However, when grammatical relations like subject and object are evaluated, parsing performance drops considerably (Green et al., 2009).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We insisted that each structure dominates a bounded number of (not necessarily adjacent) substrings.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Unbounded dependencies between branches are not possible in such a system.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The ATB has a much higher fraction of nuclei per tree, and a higher type-level error rate.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The orthographic normalization strategy we use is simple.10 In addition to removing all diacritics, we strip instances of taTweel J=J4.i, collapse variants of alif to bare alif,11 and map Ara bic punctuation characters to their Latin equivalents.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Other kinds of productive word classes, such as company names, abbreviations (termed fijsuolxie3 in Mandarin), and place names can easily be 20 Note that 7 in E 7 is normally pronounced as leO, but as part of a resultative it is liao3..","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"As we shall see, most of the linked sets are paraphrases.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In a fully supervised setting, the task is to learn a function f such that for all i = 1...m, f (xi,i, 12,i) = yz.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"gaolgaolxing4xing4 'happily' In the particular form of A-not-A reduplication illustrated in (3a), the first syllable of the verb is copied, and the negative markerbu4 'not' is inserted between the copy and the full verb.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Our results outperform strong unsupervised baselines as well as approaches that rely on direct projections, and bridge the gap between purely supervised and unsupervised POS tagging models.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Tokens were converted to vocabulary identifiers in advance and state was carried from each query to the next.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The property of semilinearity is concerned only with the occurrence of symbols in strings and not their order.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Another question that remains unanswered is to what extent the linguistic information he considers can be handled-or at least approximated-by finite-state language models, and therefore could be directly interfaced with the segmentation model that we have presented in this paper.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This range is collapsed to a number of buckets, typically by taking the hash modulo the number of buckets.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"To establish a soft correspondence between the two languages, we use a second similarity function, which leverages standard unsupervised word alignment statistics (\u00a73.3).3 Since we have no labeled foreign data, our goal is to project syntactic information from the English side to the foreign side.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"But the city name Sharm Al- Sheikh is also iDafa, hence the possibility for the incorrect annotation in (b).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The F- measure score increased for both domains, reflecting a substantial increase in recall with a small decrease in precision.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Both BBN and NYU have tagged their own data to supplement the official training data.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Interpolation search formalizes the notion that one opens a dictionary near the end to find the word \u201czebra.\u201d Initially, the algorithm knows the array begins at b +\u2014 0 and ends at e +\u2014 |A|\u22121.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"This measure has the advantage of being completely automatic.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We attain these results using several optimizations: hashing, custom lookup tables, bit-level packing, and state for left-to-right query patterns.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","label":1},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"3 Techniques for automatic vocalization have been studied (Zitouni et al., 2006; Habash and Rambow, 2007).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(a) I f f fi * fi :1 }'l ij 1\u00c2\u00a7: {1M m m s h e n 3 m e 0 shi2 ho u4 wo 3 cai2 ne ng 2 ke4 fu 2 zh e4 ge 4 ku n4 w h a t ti m e I just be abl e ov er co m e thi s C L dif fic 'When will I be able to overcome this difficulty?'","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The substructures that are unanimously hypothesized by the parsers should be preserved after combination, and the combination technique should not foolishly create substructures for which there is no supporting evidence.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We compare three hash tables: our probing implementation, GCC\u2019s hash set, and Boost\u2019s8 unordered.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We received submissions from 14 groups from 11 institutions, as listed in Figure 2.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Frequency is calculated by counting the number of distinct previous tokens that each token has (e.g., if Electric Corp. is seen 3 times, and Manufacturing Corp. is seen 5 times during training, and Corp. is not seen with any other preceding tokens, then the \u00e2\u0080\u009cfrequency\u00e2\u0080\u009d of Corp. is 2).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The cost estimate, cost(i\u00c2\u00a5JJ1l.fn is computed in the obvious way by summing the negative log probabilities of i\u00c2\u00a5JJ1l.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"More formally, we start by representing the dictionary D as a Weighted Finite State Trans\u00c2\u00ad ducer (WFST) (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In Chinese text, individual characters of the script, to which we shall refer by their traditional name of hanzi,Z are written one after another with no intervening spaces; a Chinese sentence is shown in Figure 1.3 Partly as a result of this, the notion \"word\" has never played a role in Chinese philological tradition, and the idea that Chinese lacks any\u00c2\u00ad thing analogous to words in European languages has been prevalent among Western sinologists; see DeFrancis (1984).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Raj and Whittaker (2003) show that integers in a trie implementation can be compressed substantially.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(Other classes handled by the current system are discussed in Section 5.)","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Therefore we opted to use the much 7 The DempsterShafer theory assumes that one of the hypotheses in \u00ce\u00b8 is correct, so eliminating all of the hypotheses violates this assumption.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"As far as we know, no other NERs have used information from the whole document (global) as well as information within the same sentence (local) in one framework.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"After the first step towards breadth had been taken with the PoS-tagging, RST annotation, and URML conversion of the entire corpus of 170 texts12 , emphasis shifted towards depth.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The logistic function, whose outputs are in [0, 1], forces pp(s, t) <_ po(s, t).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The type-level posterior term can be computed according to, P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2) \u00e2\u0088\u009d Note that each round of sampling Ti variables takes time proportional to the size of the corpus, as with the standard token-level HMM.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"While RST (Mann, Thompson 1988) proposed that a single relation hold between adjacent text segments, SDRT (Asher, Lascarides 2003) maintains that multiple relations may hold simultaneously.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"If one is interested in translation, one would probably want to consider show up as a single dictionary word since its semantic interpretation is not trivially derivable from the meanings of show and up.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Simply assigning to each word its most frequent associated tag in a corpus achieves 94.6% accuracy on the WSJ portion of the Penn Treebank.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"\u00e2\u0080\u00a2 Bridging links: the annotator is asked to specify the type as part-whole, cause-effect (e.g., She had an accident.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Typical data structures are generalized Bloom filters that guarantee a customizable probability of returning the correct answer.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The Levenshtein distance between the automatic translation and each of the reference translations is computed, and the minimum Levenshtein distance is taken.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Second, BABAR performs reliable case resolution to identify anaphora that can be easily resolved using the lexical and syntactic heuristics described in Section 2.1.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Both implementations employ a state object, opaque to the application, that carries information from one query to the next; we discuss both further in Section 4.2.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This suggests a direct parallel to (1): where \u02dcp(s, t) is a joint empirical distribution extracted from the IN dev set using the standard procedure.2 An alternative form of linear combination is a maximum a posteriori (MAP) combination (Bacchiani et al., 2004).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"While there might be some controversy about the exact definition of such a tagset, these 12 categories cover the most frequent part-of-speech and exist in one form or another in all of the languages that we studied.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Since MUC6, BBN' s Hidden Markov Model (HMM) based IdentiFinder (Bikel et al., 1997) has achieved remarkably good performance.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"diesem 3.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"As we shall argue, the semantic class affiliation of a hanzi constitutes useful information in predicting its properties.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Performance improvements transfer to the Moses (Koehn et al., 2007), cdec (Dyer et al., 2010), and Joshua (Li et al., 2009) translation systems where our code has been integrated.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We call this approach parse hybridization.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, a person\u00e2\u0080\u0099s full name will match with just their last name (e.g., \u00e2\u0080\u009cGeorge Bush\u00e2\u0080\u009d and \u00e2\u0080\u009cBush\u00e2\u0080\u009d), and a company name will match with and without a corporate suffix (e.g., \u00e2\u0080\u009cIBM Corp.\u00e2\u0080\u009d and \u00e2\u0080\u009cIBM\u00e2\u0080\u009d).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"markBaseNP indicates these non-recursive nominal phrases.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Table 2 shows results for both settings and all methods described in sections 2 and 3.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The linear LM (lin lm), TM (lin tm) and MAP TM (map tm) used with non-adapted counterparts perform in all cases slightly worse than the log-linear combination, which adapts both LM and TM components.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Two common cases are the attribu tive adjective and the process nominal _; maSdar, which can have a verbal reading.4 At tributive adjectives are hard because they are or- thographically identical to nominals; they are inflected for gender, number, case, and definiteness.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Our open-source (LGPL) implementation is also available for download as a standalone package with minimal (POSIX and g++) dependencies.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Thus an explicit assumption about the redundancy of the features \u2014 that either the spelling or context alone should be sufficient to build a classifier \u2014 has been built into the algorithm.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"It was filtered to retain the top 30 translations for each source phrase using the TM part of the current log-linear model.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"English parsing evaluations usually report results on sentences up to length 40.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Indeed there are several open issues.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"An important aspect of the DempsterShafer model is that it operates on sets of hypotheses.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Table 5 shows the overall parsing accuracy attained with the three different encoding schemes, compared to the baseline (no special arc labels) and to training directly on non-projective dependency graphs.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Note that Wang, Li, and Chang's.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Although we have not done any experiments on other languages, this way of using global features from a whole document should be applicable to other languages.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Compared with SRILM, IRSTLM adds several features: lower memory consumption, a binary file format with memory mapping, caching to increase speed, and quantization.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Discriminative Instance Weighting for Domain Adaptation in Statistical Machine Translation","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"0 X u} \"' o; .2 X X><X X XX X X X X X X x X X X X X x X V X X X X .;t'*- XXX:OX X X X X X X 9 x X X XX XX X X X X X X X XXX:< X X>O<XX>!KXX XI<>< \u00c2\u00bbC X X XX :X: X X \"' X X XX >OO<X>D<XIK X X X X X X --XX\u00c2\u00bb: XXX X X\u00c2\u00bbC X X\u00c2\u00abX...C:XXX X Xll< X X ><XX>IIC:liiC:oiiiiCI--8!X:liiOC!I!S8K X X X 10 100 1000 10000 log(F)_base: R\"2=0.20 (p < 0.005) X 100000 Figure 6 Plot of log frequency of base noun, against log frequency of plural nouns.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"From the point of view of recognition, independent paths in the derivation structures suggests that a top-down parser (for example) can work on each branch independently, which may lead to efficient parsing using an algorithm based on the Divide and Conquer technique.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Chunking is not enough to find such relationships.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"This is true of the widely used link grammar parser for English (Sleator and Temperley, 1993), which uses a dependency grammar of sorts, the probabilistic dependency parser of Eisner (1996), and more recently proposed deterministic dependency parsers (Yamada and Matsumoto, 2003; Nivre et al., 2004).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Foreign names are usually transliterated using hanzi whose sequential pronunciation mimics the source language pronunciation of the name.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","label":1},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"JA DE G O L D G R AS S SI C K NE SS DE AT H R A T 14.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The confidence level is then used as the belief value for the knowledge source.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Because many foreign word types are not aligned to an English word (see Table 3), and we do not run label propagation on the foreign side, we expect the projected information to have less coverage.","label":0},{"source_text":"There is no global pruning.","target_text":"The sentence length probability p(JjI) is omitted without any loss in performance.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"We will briefly discuss this point in Section 3.1.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"It is not easy to make a clear definition of \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"In the small window on the left, search queries can be entered, here one for an NP that has been annotated on the co-reference layer as bridging.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We respond to this on the one hand with a format for its underspecification (see 2.4) and on the other hand with an additional level of annotation that attends only to connectives and their scopes (see 2.5), which is intended as an intermediate step on the long road towards a systematic and objective treatment of rhetorical structure.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Next we show that the ATB is similar to other tree- banks in gross statistical terms, but that annotation consistency remains low relative to English (\u00c2\u00a73).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This variant is tested in the experiments below.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Modifying the Berkeley parser for Arabic is straightforward.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For each pair of judges, consider one judge as the standard,.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The computing time is low, since no reordering is carried out.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Here, $ is the sentence boundary symbol, which is thought to be at position 0 in the target sentence.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"For each terminal, the Leaf Ancestor metric extracts the shortest path to the root.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We thank United Informatics for providing us with our corpus of Chinese text, and BDC for the 'Behavior ChineseEnglish Electronic Dictionary.'","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"None of the models we have presented utilize features associated with a particular constituent (i.e. the label, span, parent label, etc.) to influence parser preference.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"2 Chinese ?l* han4zi4 'Chinese character'; this is the same word as Japanese kanji..","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(Yarowsky 95) describes an algorithm for word-sense disambiguation that exploits redundancy in contextual features, and gives impressive performance.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"From here on we will refer to the named-entity string itself as the spelling of the entity, and the contextual predicate as the context.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Only IRSTLM does not support threading.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"To initialize the graph for label propagation we use a supervised English tagger to label the English side of the bitext.7 We then simply count the individual labels of the English tokens and normalize the counts to produce tag distributions over English word types.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Thus in an English sentence such as I'm going to show up at the ACL one would reasonably conjecture that there are eight words separated by seven spaces.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Our system does not currently make use of titles, but it would be straightforward to do so within the finite-state framework that we propose.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Supervised methods have been applied quite successfully to the full MUC named-entity task (Bikel et al. 97).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For the English RST-annotated corpus that is made available via LDC, his corresponding result is 62%.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Of these cases, 38 were temporal expressions (either a day of the week or month of the year).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Also, the method of using keywords rules out phrases which don\u00e2\u0080\u0099t contain popular words in the domain.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"An example is in (i), where the system fails to group t;,f;?\"$?t!: lin2yang2gang3 as a name, because all three hanzi can in principle be separate words (t;,f; lin2 'wood';?\"$ yang2 'ocean'; ?t!; gang3 'harbor').","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"It is difficult when IN and OUT are dissimilar, as they are in the cases we study.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"In future work we plan to try this approach with more competitive SMT systems, and to extend instance weighting to other standard SMT components such as the LM, lexical phrase weights, and lexicalized distortion.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We can use the preceding linguistic and annotation insights to build a manually annotated Arabic grammar in the manner of Klein and Manning (2003).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"1 is given in Fig.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The procedure using the tagged sentences to discover paraphrases takes about one hour on a 2GHz Pentium 4 PC with 1GB of memory.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"REL+VB) (cf.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If the token is the first word of a sentence, then this feature is set to 1.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Lexical rules are estimated in a similar manner.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"SEGTok(noH) is the segmentation accuracy ignoring mistakes involving the implicit definite article h.11 To evaluate our performance on the tagging task we report CPOS and FPOS corresponding to coarse- and fine-grained PoS tagging results (F1) measure.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"If the log backoff of wnf is also zero (it may not be in filtered models), then wf should be omitted from the state.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"From now on all lattice arcs are tagged segments and the assignment of probability P(p \u2014* (s, p)) to lattice arcs proceeds as usual.4 A rather pathological case is when our lexical heuristics prune away all segmentation possibilities and we remain with an empty lattice.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In order to facilitate this task, we extend the set of arc labels to encode information about lifting operations.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"27 80.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"During coreference resolution, BABAR checks (1) whether the anaphor is among the lexical expectations for the caseframe that extracts the candidate antecedent, and (2) whether the candidate is among the lexical expectations for the caseframe that extracts the anaphor.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"6 Results and Analysis.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Graph construction for structured prediction problems such as POS tagging is non-trivial: on the one hand, using individual words as the vertices throws away the context necessary for disambiguation; on the other hand, it is unclear how to define (sequence) similarity if the vertices correspond to entire sentences.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"There has also been work using a bootstrap- ping approach [Brin 98; Agichtein and Gravano 00; Ravichandran and Hovy 02].","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Each distance in the traveling salesman problem now corresponds to the negative logarithm of the product of the translation, alignment and language model probabilities.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"2.2 Syntactic structure.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"kann 7.nicht 8.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The final block in table 2 shows models trained on feature subsets and on the SVM feature described in 3.4.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thus, the ATM has no more than 6km&quot; + 1 work tapes, where km&quot; is the maximum number of substrings spanned by a derived structure.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"While the paper mentioned a sorted variant, code was never released.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"For illustration, an English translation of one of the commentaries is given in Figure 1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Particular instances of relations are associated with goodness scores.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"IdentiFinder ' 99' s results are considerably better than IdentiFinder ' 97' s. IdentiFinder' s performance in MUC7 is published in (Miller et al., 1998).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Tokens tagged as PUNC are not discarded unless they consist entirely of punctuation.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"A dynamic programming algorithm is then used to select the sequence of word classes with the highest probability.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Previous work that involves the gathering of information from the whole document often uses a secondary classifier, which corrects the mistakes of a primary sentence- based classifier.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Many researchers have developed coreference resolvers, so we will only discuss the methods that are most closely related to BABAR.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The ATB is disadvantaged by having fewer trees with longer average 5 LDC A-E catalog numbers: LDC2008E61 (ATBp1v4), LDC2008E62 (ATBp2v3), and LDC2008E22 (ATBp3v3.1).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Table 1 shows results of the benchmark.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In this paper, we will propose an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"01 75.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Statistical NERs usually find the sequence of tags that maximizes the probability , where is the sequence of words in a sentence, and is the sequence of named-entity tags assigned to the words in . Attempts have been made to use global information (e.g., the same named entity occurring in different sentences of the same document), but they usually consist of incorporating an additional classifier, which tries to correct the errors in the output of a first NER (Mikheev et al., 1998; Borthwick, 1999).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In the case of adverbial reduplication illustrated in (3b) an adjective of the form AB is reduplicated as AABB.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Once we figure out the important word (e.g. keyword), we believe we can capture the meaning of the phrase by the keyword.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The second model (+PRIOR) utilizes the independent prior over type-level tag assignments P (T |\u00cf\u0088).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"orthographic words are thus only a starting point for further analysis and can only be regarded as a useful hint at the desired division of the sentence into words.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The developer explained that the loading process requires extra memory that it then frees. eBased on the ratio to SRI\u2019s speed reported in Guthrie and Hepple (2010) under different conditions.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We conjecture that this trend may continue by incorporating additional information, e.g., three-dimensional models as proposed by Tsarfaty and Sima\u2019an (2007).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The task can be considered to be one component of the MUC (MUC-6, 1995) named entity task (the other task is that of segmentation, i.e., pulling possible people, places and locations from text before sending them to the classifier).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We hope that this will allow practitioners to apply our approach directly to languages for which no resources are available.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"In the initial release of the ATB, inter-annotator agreement was inferior to other LDC treebanks (Maamouri et al., 2008).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In IdentiFinder, there is a priority in the feature assignment, such that if one feature is used for a token, another feature lower in priority will not be used.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Of these cases, 38 were temporal expressions (either a day of the week or month of the year).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"From here on we will refer to the named-entity string itself as the spelling of the entity, and the contextual predicate as the context.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This is a simple and effective alternative to setting weights discriminatively to maximize a metric such as BLEU.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"All three curves remain steep at the maximum training set size of 18818 trees.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Our implementation permits jumping to any n-gram of any length with a single lookup; this appears to be unique among language model implementations.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"In this work, we take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","label":1},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"43 80.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Figure 3 shows examples of semantic expectations that were learned.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Lattice parsing (Chappelier et al., 1999) is an alternative to a pipeline that prevents cascading errors by placing all segmentation options into the parse chart.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Daum\u00b4e (2007) applies a related idea in a simpler way, by splitting features into general and domain-specific versions.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Previous work that involves the gathering of information from the whole document often uses a secondary classifier, which corrects the mistakes of a primary sentence- based classifier.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Similarly, there is no compelling evidence that either of the syllables of f.ifflll binllang2 'betelnut' represents a morpheme, since neither can occur in any context without the other: more likely fjfflll binllang2 is a disyllabic morpheme.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Finally, since non-projective constructions often involve long-distance dependencies, the problem is closely related to the recovery of empty categories and non-local dependencies in constituency-based parsing (Johnson, 2002; Dienes and Dubey, 2003; Jijkoun and de Rijke, 2004; Cahill et al., 2004; Levy and Manning, 2004; Campbell, 2004).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In the maximum entropy framework, there is no such constraint.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"termined by the category of the word that follows it.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We currently simulate this by crafting a WCFG and feeding it to BitPar.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Borth 2MUC data can be obtained from the Linguistic Data Consortium: http:\/\/www.ldc.upenn.edu 3Training data for IdentiFinder is actually given in words (i.e., 650K & 790K words), rather than tokens Table 6: Comparison of results for MUC7 wick (1999) reported using dictionaries of person first names, corporate names and suffixes, colleges and universities, dates and times, state abbreviations, and world regions.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Many packages perform language model queries.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The second weakness is purely conceptual, and probably does not affect the per\u00c2\u00ad formance of the model.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"As can be seen from the last column in Table 1, both Head and Head+Path may theoretically lead to a quadratic increase in the number of distinct arc labels (Head+Path being worse than Head only by a constant factor), while the increase is only linear in the case of Path.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In order to solve this problem, a parse tree is needed to understand that \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is not the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"annotation guidelines that tell annotators what to do in case of doubt.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"of Articles No.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Annotation consistency is important in any supervised learning task.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The POS distributions over the foreign trigram types are used as features to learn a better unsupervised POS tagger (\u00a75).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The resulting parsers surpass the best previously published performance results for the Penn Treebank.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(4) gives In order to minimize Zt, at each iteration the final algorithm should choose the weak hypothesis (i.e., a feature xt) which has values for W+ and W_ that minimize Equ.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"0750271 and by the DARPA GALE program.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For speed, we plan to implement the direct-mapped cache from BerkeleyLM.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, there is again local grammatical information that should favor the split in the case of (1a): both .ma3 'horse' and .ma3 lu4 are nouns, but only .ma3 is consistent with the classifier pil, the classifier for horses.21 By a similar argument, the preference for not splitting , lm could be strengthened in (lb) by the observation that the classifier 1'1* tiao2 is consistent with long or winding objects like , lm ma3lu4 'road' but not with,ma3 'horse.'","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"We use MMAX for this annotation as well.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Sentence pairs are the natural instances for SMT, but sentences often contain a mix of domain-specific and general language.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Supervised methods have been applied quite successfully to the full MUC named-entity task (Bikel et al. 97).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"(1998) have also used a maximum entropy classifier that uses already tagged entities to help tag other entities.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This supports our main thesis that decisions taken by single, improved, grammar are beneficial for both tasks.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"It is a relatively frequent word in the domain, but it can be used in different extraction scenarios.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The links can solve the problem.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"As can be seen from the last column in Table 1, both Head and Head+Path may theoretically lead to a quadratic increase in the number of distinct arc labels (Head+Path being worse than Head only by a constant factor), while the increase is only linear in the case of Path.","label":0},{"source_text":"There is no global pruning.","target_text":"We have tested the translation system on the Verbmobil task (Wahlster 1993).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This fully generative model caters for real interaction between the syntactic and morphological levels as a part of a single coherent process.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The following two error criteria are used in our experiments: mWER: multi-reference WER: We use the Levenshtein distance between the automatic translation and several reference translations as a measure of the translation errors.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Formally, we define dependency graphs as follows: 3.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"pronunciation depends upon word affiliation: tfJ is pronounced deO when it is a prenominal modification marker, but di4 in the word \u00c2\u00a7tfJ mu4di4 'goal'; fl; is normally ganl 'dry,' but qian2 in a person's given name.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"TAG's can be shown to be equivalent to this restricted system.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The model is often further restricted so that each source word is assigned to exactly one target word (Brown et al., 1993; Ney et al., 2000).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Approaches differ in the algorithms used for scoring and selecting the best path, as well as in the amount of contextual information used in the scoring process.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Papers that use this method or minor variants thereof include Liang (1986), Li et al.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"All features were conjoined with the state z.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The points enumerated above are particularly related to ITS, but analogous arguments can easily be given for other applications; see for example Wu and Tseng's (1993) discussion of the role of segmentation in information retrieval.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The domains are the linguistic spans that are to receive an IS-partitioning, and the units are the (smaller) spans that can play a role as a constituent of such a partitioning.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For each pair we also record the context, i.e. the phrase between the two NEs (Step1).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"\u00e2\u0080\u009cThe gun\u00e2\u0080\u009d will be extracted by the caseframe \u00e2\u0080\u009cfired <patient>\u00e2\u0080\u009d.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"There are two possible reasons: (1) the knowledge sources are resolving different cases of anaphora, and (2) the knowledge sources provide multiple pieces of evidence in support of (or against) a candidate, thereby acting synergistically to push the DempsterShafer model over the belief threshold in favor of a single candidate.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"IdentiFinder ' 99' s results are considerably better than IdentiFinder ' 97' s. IdentiFinder' s performance in MUC7 is published in (Miller et al., 1998).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The idea of distinguishing between general and domain-specific examples is due to Daum\u00b4e and Marcu (2006), who used a maximum-entropy model with latent variables to capture the degree of specificity.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"By aggregating the POS labels of the English tokens to types, we can generate label distributions for the English vertices.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"For all grammars, we use fine-grained PoS tags indicating various morphological features annotated therein.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"This group contains a large number of features (one for each token string present in the training data).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"(2009).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Named Entity Recognition: A Maximum Entropy Approach Using Global Information","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"This line of work has been motivated by empirical findings that the standard EM-learned unsupervised HMM does not exhibit sufficient word tag sparsity.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Memory mapping also allows the same model to be shared across processes on the same machine.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The confidence intervals are computed by bootstrap resampling for BLEU, and by standard significance testing for the manual scores, as described earlier in the paper.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"shortest match at each point.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Another question that remains unanswered is to what extent the linguistic information he considers can be handled-or at least approximated-by finite-state language models, and therefore could be directly interfaced with the segmentation model that we have presented in this paper.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Training examples are generated automatically by identifying noun phrases that can be easily resolved with their antecedents using lexical and syntactic heuristics.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, if X and Y are coreferent, then both X and Y are considered to co-occur with the caseframe that extracts X as well as the caseframe that extracts Y. We will refer to the set of nouns that co-occur with a caseframe as the lexical expectations of the case- frame.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Out-of-Vocabulary: We derived a lexicon list from WordNet 1.6, and words that are not found in this list have a feature out-of-vocabulary set to 1.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The Berkeley parser gives state-of-the-art performance for all metrics.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"This design does not guarantee \u00e2\u0080\u009cstructural zeros,\u00e2\u0080\u009d but biases towards sparsity.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"This feature includes named entities, which the ATB marks with a flat NP node dominating an arbitrary number of NNP pre-terminal daughters (Figure 2).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Hearst 92) describes a method for extracting hyponyms from a corpus (pairs of words in &quot;isa&quot; relations).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The tokens w are generated by token-level tags t from an HMM parameterized by the lexicon structure.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"It falls short of the \u201cProjection\u201d baseline for German, but is statistically indistinguishable in terms of accuracy.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We use label propagation in two stages to generate soft labels on all the vertices in the graph.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"(a) Of the high frequency phrasal categories, ADJP and SBAR are the hardest to parse.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Our smoothing procedure takes into account all the aforementioned aspects and works as follows.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The CoBoost algorithm described above divides the function Zco into two parts: Zco = 40 + 40.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Each trie node is individually allocated and full 64-bit pointers are used to find them, wasting memory.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The details are given in (Och and Ney, 2000).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"A greedy algorithm (or maximum-matching algorithm), GR: proceed through the sentence, taking the longest match with a dictionary entry at each point.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The model described here thus demonstrates great potential for use in widespread applications.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"And indeed, converging on annotation guidelines is even more difficult than it is with co-reference.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Finally, quite a few hanzi are homographs, meaning that they may be pronounced in several different ways, and in extreme cases apparently represent different morphemes: The prenominal modifi\u00c2\u00ad cation marker eg deO is presumably a different morpheme from the second morpheme of \u00c2\u00a7eg mu4di4, even though they are written the same way.4 The second point, which will be relevant in the discussion of personal names in Section 4.4, relates to the internal structure of hanzi.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For terrorism, BABAR generated 5,078 resolutions: 2,386 from lexical seeding and 2,692 from syntactic seeding.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The web-based Annis imports data in a variety of XML formats and tagsets and displays it in a tier-orientedway (optionally, trees can be drawn more ele gantly in a separate window).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We have checked if there are similar verbs in other major domains, but this was the only one.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"86 78.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Our new approach, which is called quasi-monotone search, processes the source sentence monotonically, while explicitly taking into account the positions of the German verbgroup.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Note that because we extracted only high-confidence alignments, many foreign vertices will not be connected to any English vertices.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"These are written to the state s(wn1) and returned so that they can be used for the following query.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"However, those methods need initial seeds, so the relation between entities has to be known in advance.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We present several variations for the lexical component P (T , W |\u00cf\u0088), each adding more complex pa rameterizations.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We use the default inference parameters.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The major problem for our seg\u00c2\u00ad menter, as for all segmenters, remains the problem of unknown words (see Fung and Wu [1994]).","label":0},{"source_text":"There is no global pruning.","target_text":"The translation direction is from German to English.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Formally, let el (62) be the number of classification errors of the first (second) learner on the training data, and let Eco be the number of unlabeled examples on which the two classifiers disagree.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"A corpus of German newspaper commentaries has been assembled and annotated with different information (and currently, to different degrees): part-of-speech, syntax, rhetorical structure, connectives, co-reference, and information structure.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"(2009), who also incorporate a sparsity constraint, but does via altering the model objective using posterior regularization.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Clearly it is possible to write a rule that states that if an analysis Modal+ Verb is available, then that is to be preferred over Noun+ Verb: such a rule could be stated in terms of (finite-state) local grammars in the sense of Mohri (1993).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Our motivation for using DempsterShafer is that it provides a well-principled framework for combining evidence from multiple sources with respect to competing hypotheses.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The relevant variables are the set of token-level tags that appear before and after each instance of the ith word type; we denote these context pairs with the set {(tb, ta)} and they are contained in t(\u00e2\u0088\u0092i).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Kollege.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(3)), with one term for each classifier.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Each distance in the traveling salesman problem now corresponds to the negative logarithm of the product of the translation, alignment and language model probabilities.","label":0},{"source_text":"There is no global pruning.","target_text":"The final score is obtained from: max e;e0 j2fJ\u00f4\u0080\u0080\u0080L;;Jg p($je; e0) Qe0 (e; I; f1; ; Jg; j); where p($je; e0) denotes the trigram language model, which predicts the sentence boundary $ at the end of the target sentence.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"An error count of 0:0 is assigned to a perfect translation, and an error count of 1:0 is assigned to a semantically and syntactically wrong translation.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"When the parser is trained on the transformed data, it will ideally learn not only to construct projective dependency structures but also to assign arc labels that encode information about lifts.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Our analysis identifies three key factors driving our performance gain: 1) selecting a model structure which directly encodes tag sparsity, 2) a type-level prior on tag assignments, and 3) a straightforward na\u00c2\u00a8\u00c4\u00b1veBayes approach to incorporate features.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"83 77.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Table 2 shows these similarity measures.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As indicated in Figure 1(c), apart from this correct analysis, there is also the analysis taking B ri4 as a word (e.g., a common abbreviation for Japan), along with X:\u00c2\u00a5 wen2zhangl 'essay\/ and f!!.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"suffixes (e.g., \u00ef\u00bf\u00bd=) Other notable parameters are second order vertical Markovization and marking of unary rules.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The edge weights between the foreign language trigrams are computed using a co-occurence based similarity function, designed to indicate how syntactically similar the middle words of the connected trigrams are (\u00a73.2).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This is a standard adaptation problem for SMT.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In order to facilitate this task, we extend the set of arc labels to encode information about lifting operations.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Because we don\u2019t have a separate development set, we used the training set to select among them and found 0.2 to work slightly better than 0.1 and 0.3.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This implies, therefore, that a major factor in the performance of a Chinese segmenter is the quality of the base dictionary, and this is probably a more important factor-from the point of view of performance alone-than the particular computational methods used.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Table 5: Effect of the beam threshold on the number of search errors (147 sentences).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The types of patterns produced by AutoSlog are outlined in (Riloff, 1996).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Our assumption is that caseframes that co-occur in resolutions often have a 2 This normalization is performed syntactically without semantics, so the agent and patient roles are not guaranteed to hold, but they usually do in practice.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":".","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"4.3 Morphological Analysis.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The most frequently occurring last words of organization names in cslist are compiled into a list of corporate suffixes, Corporate-Suffix- List.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In the denomi 11 We have two such lists, one containing about 17,000 full names, and another containing frequencies of.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"47 78.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"9 61.0 44.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Following the convention presented in earlier sections, we assume that each example is an instance pair of the from (xi ,i, x2,) where xj,, E 2x3 , j E 2}.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Note that these observa sider suffix features, capitalization features, punctuation, and digit features.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"64 94.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Preferences for constituent order (especially in languages with relatively free word order) often belong to this group.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Of course, we are primarily interested in applying our techniques to languages for which no labeled resources are available.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"We assume that the goal in dependency parsing is to construct a labeled dependency graph of the kind depicted in Figure 1.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"BerkeleyLM uses states to optimistically search for longer n-gram matches first and must perform twice as many random accesses to retrieve backoff information.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Process statistics are already collected by the kernel (and printing them has no meaningful impact on performance).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"of Arabic.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Automatic paraphrase discovery is an important but challenging task.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"A crucial difference is that the number of parameters is greatly reduced as is the number of variables that are sampled during each iteration.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This is an issue that we have not addressed at the current stage of our research.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"68 96.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Unfortunately, there is no standard corpus of Chinese texts, tagged with either single or multiple human judgments, with which one can compare performance of various methods.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"4 69.0 51.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"This result suggests that all of contextual role KSs can provide useful information for resolving anaphora.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Each decision determines the inclusion or exclusion of a candidate constituent.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"It is well-known that constituency parsing models designed for English often do not generalize easily to other languages and treebanks.1 Explanations for this phenomenon have included the relative informativeness of lexicalization (Dubey and Keller, 2003; Arun and Keller, 2005), insensitivity to morphology (Cowan and Collins, 2005; Tsarfaty and Sima\u00e2\u0080\u0099an, 2008), and the effect of variable word order (Collins et al., 1999).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Note that since all English vertices were extracted from the parallel text, we will have an initial label distribution for all vertices in Ve.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"This fact annoyed especially his dog...).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For a sequence of hanzi that is a possible name, we wish to assign a probability to that sequence qua name.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"We have presented a new method for non-projective dependency parsing, based on a combination of data-driven projective dependency parsing and graph transformation techniques.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Also, expanding on the techniques for the automatic generation of extraction patterns (Riloff 96; Sudo 03) using our method, the extraction patterns which have the same meaning can be automatically linked, enabling us to produce the final table fully automatically.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"While it is possible to derive a closed form solution for this convex objective function, it would require the inversion of a matrix of order |Vf|.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"In graph-based learning approaches one constructs a graph whose vertices are labeled and unlabeled examples, and whose weighted edges encode the degree to which the examples they link have the same label (Zhu et al., 2003).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We hypothesize that modeling morphological information will greatly constrain the set of possible tags, thereby further refining the representation of the tag lexicon.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"A contextual rule considers words surrounding the string in the sentence in which it appears (e.g., a rule that any proper name modified by an appositive whose head is president is a person).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Lexical and Morphological Ambiguity The rich morphological processes for deriving Hebrew stems give rise to a high degree of ambiguity for Hebrew space-delimited tokens.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"2.2.3 Lexical Caseframe Expectations The second type of contextual role knowledge learned by BABAR is Lexical Caseframe Expectations, which are used by the CFLex knowledge source.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Little attention, however, has been paid to the structural descriptions that these formalisms can assign to strings, i.e. their strong generative capacity.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"(Charniak et al., 1996).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Lexicon Feature of Previous and Next Token: The string of the previous token and the next token is used with the initCaps information of . If has initCaps, then a feature (initCaps, ) is set to 1.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For example, in predicting if a word belongs to a word class, is either true or false, and refers to the surrounding context: if = true, previous word = the otherwise The parameters are estimated by a procedure called Generalized Iterative Scaling (GIS) (Darroch and Ratcliff, 1972).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Similarly Figures 1 and 2 show how the isolated constituent precision varies by sentence length and the size of the span of the hypothesized constituent.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"AdaBoost is given access to a weak learning algorithm, which accepts as input the training examples, along with a distribution over the instances.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The natural baseline approach is to concatenate data from IN and OUT.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Finally, the statistical method fails to correctly group hanzi in cases where the individual hanzi comprising the name are listed in the dictionary as being relatively high-frequency single-hanzi words.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We can only compare with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"This paper proposes a simple and effective tagging method that directly models tag sparsity and other distributional properties of valid POS tag assignments.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The two new terms force the two classifiers to agree, as much as possible, on the unlabeled examples.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Mikheev et al.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"(Hartmann 1984), for example, used the term Reliefgebung to characterize the distibution of main and minor information in texts (similar to the notion of nuclearity in RST).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Our second point of comparison is with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We run the baseline Moses system for the French-English track of the 2011 Workshop on Machine Translation,9 translating the 3003-sentence test set.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"(Again, the goal of also in structural features.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"With some minor API changes, namely returning the length of the n-gram matched, it could also be faster\u2014though this would be at the expense of an optimization we explain in Section 4.1.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The first setting uses the European Medicines Agency (EMEA) corpus (Tiedemann, 2009) as IN, and the Europarl (EP) corpus (www.statmt.org\/europarl) as OUT, for English\/French translation in both directions.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"In this paper we have proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Using these states, we define partial hypothesis extensions, which are of the following type: (S0;C n fjg; j0) !","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"A crucial difference is that the number of parameters is greatly reduced as is the number of variables that are sampled during each iteration.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We offer a state function s(wn1) = wn\ufffd where substring wn\ufffd is guaranteed to extend (to the right) in the same way that wn1 does for purposes of language modeling.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Further, we report current resident memory and peak virtual memory because these are the most applicable statistics provided by the kernel.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In the following, we assume that this word joining has been carried out.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The algorithm builds two classifiers iteratively: each iteration involves minimization of a continuously differential function which bounds the number of examples on which the two classifiers disagree.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Evaluation metrics used are Attachment Score (AS), i.e. the proportion of tokens that are attached to the correct head, and Exact Match (EM), i.e. the proportion of sentences for which the dependency graph exactly matches the gold standard.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"8 1 8.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Restrictions We compare our new approach with the word reordering used in the IBM translation approach (Berger et al., 1996).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"For even larger models, storing counts (Talbot and Osborne, 2007; Pauls and Klein, 2011; Guthrie and Hepple, 2010) is a possibility.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The commentaries in PCC are all of roughly the same length, ranging from 8 to 10 sentences.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The knowledge base then can be tested for its relation-inference capabilities on the basis of full-blown co-reference information.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In the case of, the most common usage is as an adverb with the pronunciation jiangl, so that variant is assigned the estimated cost of 5.98, and a high cost is assigned to nominal usage with the pronunciation jiang4.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Once we figure out the important word (e.g. keyword), we believe we can capture the meaning of the phrase by the keyword.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The difference between the featureless model (+PRIOR) and our full model (+FEATS) is 13.6% and 7.7% average error reduction on best and median settings respectively.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For the experiments, we used four newswire corpora, the Los Angeles Times\/Washington Post, The New York Times, Reuters and the Wall Street Journal, all published in 1995.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In this paper, we show how non-projective dependency parsing can be achieved by combining a datadriven projective parser with special graph transformation techniques.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Additionally, for a given coverage set, at most 250 different hypotheses are kept during the search process, and the number of different words to be hypothesized by a source word is limited.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We refer to a segment and its assigned PoS tag as a lexeme, and so analyses are in fact sequences of lexemes.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In this case nonalpha is the string formed by removing all upper\/lower case letters from the spelling (e.g., for Thomas E. Petry nonalpha= .","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Gazdar (1985) argues this is the appropriate analysis of unbounded dependencies in the hypothetical Scandinavian language Norwedish.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"If there is a frequent multi-word sequence in a domain, we could use it as a keyword candidate.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"On several languages, we report performance exceeding that of state-of-the art systems.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Our code has been publicly available and intergrated into Moses since October 2010.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"As lower frequency examples include noise, we set a threshold that an NE category pair should appear at least 5 times to be considered and an NE instance pair should appear at least twice to be considered.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"\u00c2\u00a3 : _ADV: 5.88 If:!","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Since Dane\u00cb\u0087s\u00e2\u0080\u0099 proposals of \u00e2\u0080\u0098thematic development patterns\u00e2\u0080\u0099, a few suggestions have been made as to the existence of a level of discourse structure that would predict the information structure of sentences within texts.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Set the decision list to include all rules whose (smoothed) strength is above some threshold Pmin.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, a story can mention \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d, \u00e2\u0080\u009cthe White House\u00e2\u0080\u009d, or \u00e2\u0080\u009cthe weather\u00e2\u0080\u009d without any prior referent in the story.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Virginia) and -sia are normally transliterated as fbSi!","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"3 68.4 49.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We do not show the numbers for the Bayes models in Table 2 because the parameters involved were established using this set.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Analogous to the sharing of stacks in IC's, Lexical-Functional Grammar's (LFG's) use the unification of unbounded hierarchical structures.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We adopted this state-of-the-art model because it makes it easy to experiment with various ways of incorporating our novel constraint feature into the log-linear emission model.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"8 1 2.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"4.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We define the following function: If Zco is small, then it follows that the two classifiers must have a low error rate on the labeled examples, and that they also must give the same label on a large number of unlabeled instances.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In the second part of the experiment, we applied the inverse transformation based on breadth-first search under the three different encoding schemes.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The token-level term is similar to the standard HMM sampling equations found in Johnson (2007).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Following Sproat and Shih (1990), performance for Chinese segmentation systems is generally reported in terms of the dual measures of precision and recalP It is fairly standard to report precision and recall scores in the mid to high 90% range.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We retain segmentation markers\u00e2\u0080\u0094which are consistent only in the vocalized section of the treebank\u00e2\u0080\u0094to differentiate between e.g. \u00ef\u00bf\u00bd \u00e2\u0080\u009cthey\u00e2\u0080\u009d and \u00ef\u00bf\u00bd + \u00e2\u0080\u009ctheir.\u00e2\u0080\u009d Because we use the vocalized section, we must remove null pronoun markers.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"If evidence indicates that hypotheses C and D are less likely than hypotheses A and B, then probabilities are redistributed to reflect the fact that {A, B} is more likely to contain the answer than {C, D}.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We are currently exploring other methods that employ similar ideas and their formal properties.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"They contain about 200M words (25M, 110M, 40M and 19M words, respectively).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In this work we extended the AdaBoost.MH (Schapire and Singer 98) algorithm to the cotraining case.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our coreference resolver performed well in two domains, and experiments showed that each contextual role knowledge source contributed valuable information.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(4) is minimized by setting Since a feature may be present in only a few examples, W_ can be in practice very small or even 0, leading to extreme confidence values.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"keywords Step 3 Sets of phrases based on keywords Step 4 Links between sets of phrases All the contexts collected for a given domain are gathered in a bag and the TF\/ITF scores are calculated for all the words except stopwords in the bag.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"8 Volume 1, English language, 19961997, Format version 1, correction level 0 An ap ho r T e r r o r i s m R e c Pr F D i s a s t e r s R e c Pr F De f. NP s Pro no uns .43 .79 .55 .50 .72 .59 .42 .91 .58 .42 .82 .56 Tot al .46 .76 .57 .42 .87 .57 Table 2: General Knowledge Sources Table 4: Individual Performance of KSs for Terrorism Table 3: General + Contextual Role Knowledge Sources larger MUC4 and Reuters corpora.9 4.2 Experiments.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Frequency is calculated by counting the number of distinct previous tokens that each token has (e.g., if Electric Corp. is seen 3 times, and Manufacturing Corp. is seen 5 times during training, and Corp. is not seen with any other preceding tokens, then the \u00e2\u0080\u009cfrequency\u00e2\u0080\u009d of Corp. is 2).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Translation errors are reported in terms of multireference word error rate (mWER) and subjective sentence error rate (SSER).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We can now add a new weak hypothesis 14 based on a feature in X1 with a confidence value al hl and atl are chosen to minimize the function We now define, for 1 <i <n, the following virtual distribution, As before, Ztl is a normalization constant.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The algorithm builds two classifiers iteratively: each iteration involves minimization of a continuously differential function which bounds the number of examples on which the two classifiers disagree.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"TRIE uses less memory and has better locality.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"!!:\\ :yu2 e:_nc [::!!:zen3 l!f :moO t:_adv il!:shuot ,:_vb i i i 1 \u00e2\u0080\u00a2 10.03 13...","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Several methods exist to handle collisions; we use linear probing because it has less memory overhead when entries are small.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The major problem for our seg\u00c2\u00ad menter, as for all segmenters, remains the problem of unknown words (see Fung and Wu [1994]).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The model we use provides a simple framework in which to incorporate a wide variety of lexical information in a uniform way.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"1 is given in Fig.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"shows some keywords with their scores.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Figure 1: Translation of PCC sample commentary (STTS)2.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We thank Felix Hageloh (Hageloh, 2006) for providing us with this version. proposed in (Tsarfaty, 2006).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Conditioned on T , features of word types W are drawn.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In many cases, there is an even stronger restriction: over large portions of the source string, the alignment is monotone.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"A contextual role represents the role that a noun phrase plays in an event or relationship.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We resolve this problem by inserting an entry with probability set to an otherwise-invalid value (\u2212oc).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"In Table 7 we give results for several evaluation metrics.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"yields.6 But to its great advantage, it has a high ratio of non-terminals\/terminals (\u00ce\u00bc Constituents \/ \u00ce\u00bc Length).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"of Arabic.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Our code has been publicly available and intergrated into Moses since October 2010.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The Prague Dependency Treebank (PDT) consists of more than 1M words of newspaper text, annotated on three levels, the morphological, analytical and tectogrammatical levels (Haji\u02c7c, 1998).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For E(ni1s), then, we substitute a smooth S against the number of class elements.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This is especially bad with PROBING because it is based on hashing and performs random lookups, but it is not intended to be used in low-memory scenarios.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Restrictions We compare our new approach with the word reordering used in the IBM translation approach (Berger et al., 1996).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Domain adaptation is a common concern when optimizing empirical NLP applications.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Lexical-knowledge-based approaches that include statistical information generally presume that one starts with all possible segmentations of a sentence, and picks the best segmentation from the set of possible segmentations using a probabilistic or cost\u00c2\u00ad based scoring mechanism.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"This stage of label propagation results in a tag distribution ri over labels y, which encodes the proportion of times the middle word of ui E Vf aligns to English words vy tagged with label y: The second stage consists of running traditional label propagation to propagate labels from these peripheral vertices Vf\ufffd to all foreign language vertices in the graph, optimizing the following objective: 5 POS Induction After running label propagation (LP), we compute tag probabilities for foreign word types x by marginalizing the POS tag distributions of foreign trigrams ui = x\u2212 x x+ over the left and right context words: where the qi (i = 1, ... , |Vf|) are the label distributions over the foreign language vertices and \u00b5 and \u03bd are hyperparameters that we discuss in \u00a76.4.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"An anti-greedy algorithm, AG: instead of the longest match, take the.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"To lower the barrier of entrance to the competition, we provided a complete baseline MT system, along with data resources.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The resulting structural differences between tree- banks can account for relative differences in parsing performance.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"4.1 Dictionary Representation.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The Leaf Ancestor metric measures the cost of transforming guess trees to the reference (Sampson and Babarczy, 2003).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Naseem et al. (2009) and Snyder et al.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Restrictions We compare our new approach with the word reordering used in the IBM translation approach (Berger et al., 1996).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Data We use the Hebrew Treebank, (Sima\u2019an et al., 2001), provided by the knowledge center for processing Hebrew, in which sentences from the daily newspaper \u201cHa\u2019aretz\u201d are morphologically segmented and syntactically annotated.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"If there is a frequent multi-word sequence in a domain, we could use it as a keyword candidate.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If it starts with a lower case letter, and contains both upper and lower case letters, then (mixedCaps, zone) is set to 1.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"So, there is a limitation that IE can only be performed for a predefined task, like \u00e2\u0080\u009ccorporate mergers\u00e2\u0080\u009d or \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Table 1 The cost as a novel given name (second position) for hanzi from various radical classes.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"3.4 Salience-based text generation.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"This can be repeated several times to collect a list of author \/ book title pairs and expressions.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Third, all remaining anaphora are evaluated by 11 different knowledge sources: the four contextual role knowledge sources just described and seven general knowledge sources.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If is one of Monday, Tuesday, . . .","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The DL-CoTrain algorithm can be motivated as being a greedy method of satisfying the above 2 constraints.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"We then evaluate the approach in two steps.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Let 71 be a tree with root and foot labeled by X.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Memory mapping also allows the same model to be shared across processes on the same machine.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"0 Figure 5 An example of affixation: the plural affix.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In total, across all domains, we kept 13,976 phrases with keywords.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(7) is at 0 when: 1) Vi : sign(gi (xi)) = sign(g2 (xi)); 2) Ig3(xi)l oo; and 3) sign(gi (xi)) = yi for i = 1, , m. In fact, Zco provides a bound on the sum of the classification error of the labeled examples and the number of disagreements between the two classifiers on the unlabeled examples.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"ni2ya3 and @5:2 xilya3, respectively.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Full Chinese personal names are in one respect simple: they are always of the form family+given.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"x EG, has agreed to be bought by H x EG, now owned by H x H to acquire EG x H\u00e2\u0080\u0099s agreement to buy EG Three of those phrases are actually paraphrases, but sometime there could be some noise; such as the second phrase above.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"As we shall argue, the semantic class affiliation of a hanzi constitutes useful information in predicting its properties.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Table 1 briefly describes the seven syntactic heuristics used by BABAR to resolve noun phrases.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The development of automatic scoring methods is an open field of research.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"RandLM\u2019s stupid backoff variant stores counts instead of probabilities and backoffs.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"For the inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J), we drop the dependence on the target sentence length I. 2.2 Word Joining.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"(1) CEO of McCann . . .","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We measured recall (Rec), precision (Pr), and the F-measure (F) with recall and precision equally weighted.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Yarowsky-cautious does not separate the spelling and contextual features, but does have a limit on the number of rules added at each stage.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In IdentiFinder, there is a priority in the feature assignment, such that if one feature is used for a token, another feature lower in priority will not be used.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If these sets do not overlap, then the words cannot be coreferent.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Compared with the widely- SRILM, our is 2.4 times as fast while using 57% of the mem- The structure is a trie with bit-level packing, sorted records, interpolation search, and optional quantization aimed lower memory consumption. simultaneously uses less memory than the smallest lossless baseline and less CPU than the baseline.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Finally, as\u00c2\u00ad suming a simple bigram backoff model, we can derive the probability estimate for the particular unseen word i\u00c2\u00a51J1l.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"An edge indicates that the two features must have the same label.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Information from a sentence is sometimes insufficient to classify a name correctly.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"More recently, (Riloff and Jones 99) describe a method they term &quot;mutual bootstrapping&quot; for simultaneously constructing a lexicon and contextual extraction patterns.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We evaluated BABAR on two domains: terrorism and natural disasters.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We do not attempt to identify the types of relationships that are found.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"These enable much larger models in memory, compensating for lost accuracy.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Translation errors are reported in terms of multireference word error rate (mWER) and subjective sentence error rate (SSER).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The function A : F \u2014* C maps from the language specific fine-grained tagset F to the coarser universal tagset C and is described in detail in \u00a76.2: Note that when tx(y) = 1 the feature value is 0 and has no effect on the model, while its value is \u2212oc when tx(y) = 0 and constrains the HMM\u2019s state space.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"to explore how well we can induce POS tags using only the one-tag-per-word constraint.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Since the lattice L for a given sentence W is determined by the morphological analyzer M we have which is precisely the formula corresponding to the so-called lattice parsing familiar from speech recognition.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Queries detect the invalid probability, using the node only if it leads to a longer match.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"\u00e2\u0080\u009d, because Bush is the first word, the initial caps might be due to its position (as in \u00e2\u0080\u009cThey put a freeze on . . .","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"In the first stage, we run a single step of label propagation, which transfers the label distributions from the English vertices to the connected foreign language vertices (say, Vf\ufffd) at the periphery of the graph.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The NE tagger is a rule-based system with 140 NE categories [Sekine et al. 2004].","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"\u00ce\u00b2 is the shared hyperparameter for the tag assignment prior and word feature multinomials.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The most informative scheme, Head+Path, gives the highest scores, although with respect to Head the difference is not statistically significant, while the least informative scheme, Path \u2013 with almost the same performance on treebank transformation \u2013 is significantly lower (p < 0.01).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"It can be shown that this objective is convex in q.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We define the lattice L to be the concatenation of the lattices Li corresponding to the input words wi (s.t.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"First, we aim to explicitly characterize examples from OUT as belonging to general language or not.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In the numerator, however, the counts of ni1s are quite irregular, in\u00c2\u00ad cluding several zeros (e.g., RAT, none of whose members were seen).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, BABAR learned that agents that \u00e2\u0080\u009cassassinate\u00e2\u0080\u009d or \u00e2\u0080\u009cinvestigate a cause\u00e2\u0080\u009d are usually humans or groups (i.e., organizations).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"(Hartmann 1984), for example, used the term Reliefgebung to characterize the distibution of main and minor information in texts (similar to the notion of nuclearity in RST).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(a) 1 \u00c2\u00a7 . ;m t 7 leO z h e 4 pil m a 3 lu 4 sh an g4 bi ng 4 t h i s CL (assi fier) horse w ay on sic k A SP (ec t) 'This horse got sick on the way' (b) 1\u00c2\u00a7: . til y zhe4 tiao2 ma3lu4 hen3 shao3 this CL road very few 'Very few cars pass by this road' :$ chel jinglguo4 car pass by 2.","label":0},{"source_text":"There is no global pruning.","target_text":"Thus, the effects of spontaneous speech are present in the corpus, e.g. the syntactic structure of the sentence is rather less restricted, however the effect of speech recognition errors is not covered.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We call this approach parse hybridization.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"During coreference resolution, the caseframe network provides evidence that an anaphor and prior noun phrase might be coreferent.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Using the virtual distribution Di (i) and pseudo-labels&quot;y.,\u201e values for Wo, W\u00b1 and W_ can be calculated for each possible weak hypothesis (i.e., for each feature x E Xi); the weak hypothesis with minimal value for Wo + 2\/WW _ can be chosen as before; and the weight for this weak hypothesis at = ln ww+411:) can be calculated.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Evaluation of links A link between two sets is considered correct if the majority of phrases in both sets have the same meaning, i.e. if the link indicates paraphrase.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We reduce this to O(log log |A|) time by evenly distributing keys over their range then using interpolation search4 (Perl et al., 1978).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In the last example, the less restrictive IbmS word reordering leads to a better translation, although the QmS translation is still acceptable.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective treebanks is often neglected because of the relative scarcity of problematic constructions.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"\u201cin the shadow\u201d.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR then computes statistics over the training examples measuring the frequency with which extraction patterns and noun phrases co-occur in coreference resolutions.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"When two partial hypotheses have equal state (including that of other features), they can be recombined and thereafter efficiently handled as a single packed hypothesis.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"It is worth noting that the middle words of the Italian trigrams are nouns too, which exhibits the fact that the similarity metric connects types having the same syntactic category.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"5.1 Data Sets.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Other packages walk their respective data structures once to find wnf and again to find {b(wn\u22121 i )}f\u22121 i=1if necessary.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"40 75.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"During coreference resolution, the caseframe network provides evidence that an anaphor and prior noun phrase might be coreferent.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Learned Tag Prior (PRIOR) We next assume there exists a single prior distribution \u00cf\u0088 over tag assignments drawn from DIRICHLET(\u00ce\u00b2, K ).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The \u201cNo LP\u201d model does not outperform direct projection for German and Greek, but performs better for six out of eight languages.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Similar behavior is observed when adding features.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"If e < b then the key is not found.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Due to the dramatic fiscal situation in Brandenburg she now surprisingly withdrew legislation drafted more than a year ago, and suggested to decide on it not before 2003.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Time starts when Moses is launched and therefore includes model loading time.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"At first glance, this seems only peripherally related to our work, since the specific\/general distinction is made for features rather than instances.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For example, in predicting if a word belongs to a word class, is either true or false, and refers to the surrounding context: if = true, previous word = the otherwise The parameters are estimated by a procedure called Generalized Iterative Scaling (GIS) (Darroch and Ratcliff, 1972).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"This can be seen as a rough approximation of Yarowsky and Ngai (2001).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Domain adaptation is a common concern when optimizing empirical NLP applications.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Evaluation Metrics We report three metrics to evaluate tagging performance.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"then define the best segmentation to be the cheapest or best path in Id(I) o D* (i.e., Id(I) composed with the transitive closure of 0).6 Consider the abstract example illustrated in Figure 2.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In (2a), we want to split the two morphemes since the correct analysis is that we have the adverb :1 cai2 'just,' the modal verb neng2 'be able' and the main verb R: Hke4fu2 'overcome'; the competing analysis is, of course, that we have the noun :1 cai2neng2 'talent,' followed by }'lijke4fu2 'overcome.'","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The most accurate characterization of Chinese writing is that it is morphosyllabic (DeFrancis 1984): each hanzi represents one morpheme lexically and semantically, and one syllable phonologi\u00c2\u00ad cally.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"To check for this, we do pairwise bootstrap resampling: Again, we repeatedly sample sets of sentences, this time from both systems, and compare their BLEU scores on these sets.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For instance, the common \"suffixes,\" -nia (e.g.,.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We have developed a general approach for combining parsers when preserving the entire structure of a parse tree is important.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The counts represent portions of the approximately 44000 constituents hypothesized by the parsers in the development set.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Considerable amount of work has been done in recent years on the named entity recognition task, partly due to the Message Understanding Conferences (MUC).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The equivalence of CC's with this restriction to TAG's and HG's is, however, still an open problem.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Besides information structure, the second main goal is to enhance current models of rhetorical structure.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"6) are noisy, the results confirm that label propagation within the foreign language part of the graph adds significant quality for every language.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"But we also need an estimate of the probability for a non-occurring though possible plural form like i\u00c2\u00a5JJ1l.f, nan2gua1-men0 'pumpkins.'","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Note that the backoff model assumes that there is a positive correlation between the frequency of a singular noun and its plural.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The general idea for the knowledge- based part is to have the system use as much information as it can find at its disposal to produce a target representation as specific as possible and as underspecified as necessary.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Its correct antecedent is \u00e2\u0080\u009ca revolver\u00e2\u0080\u009d, which is extracted by the caseframe \u00e2\u0080\u009ckilled with <NP>\u00e2\u0080\u009d.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For a description of the application of AdaBoost to various NLP problems see the paper by Abney, Schapire, and Singer in this volume.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"There have been other kinds of efforts to discover paraphrase automatically from corpora.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"While sorted arrays could be used to implement the same data structure as PROBING, effectively making m = 1, we abandoned this implementation because it is slower and larger than a trie implementation.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"In this particular case, all English vertices are labeled as nouns by the supervised tagger.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Syntactic universals are a well studied concept in linguistics (Carnie, 2002; Newmeyer, 2005), and were recently used in similar form by Naseem et al. (2010) for multilingual grammar induction.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The method just described segments dictionary words, but as noted in Section 1, there are several classes of words that should be handled that are not found in a standard dictionary.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The sources of our dictionaries are listed in Table 2.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The model starts by generating a tag assignment for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Current state-of-the-art broad-coverage parsers assume a direct correspondence between the lexical items ingrained in the proposed syntactic analyses (the yields of syntactic parse-trees) and the spacedelimited tokens (henceforth, \u2018tokens\u2019) that constitute the unanalyzed surface forms (utterances).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"We use a solution to this problem similar to the one presented in (Och et al., 1999), where target words are joined during training.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We reduce this to O(log log |A|) time by evenly distributing keys over their range then using interpolation search4 (Perl et al., 1978).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"La ng ua ge 1T W + P RI O R + F E A T S E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 2 1.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Proper names that match are resolved with each other.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"computing the recall of the other's judgments relative to this standard.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"However, the point of RandLM is to scale to even larger data, compensating for this loss in quality.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The first value reports resident size after loading; the second is the gap between post-loading resident memory and peak virtual memory.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"For our demonstration system, we typically use the pruning threshold t0 = 5:0 to speed up the search by a factor 5 while allowing for a small degradation in translation accuracy.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Process nominals name the action of the transitive or ditransitive verb from which they derive.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We did have a problem classifying some modified noun phrases where the modified phrase does not represent a qualified or restricted form of the head, like \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, as these are both represented by the keyword \u00e2\u0080\u009cchairman\u00e2\u0080\u009d.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"36 79.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The transition from f, to a final state transduces c to the grammatical tag PL with cost cost(unseen(f,)): cost(i\u00c2\u00a5JJ1l.ir,) == cost(i\u00c2\u00a5JJ1l.)","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We tried two versions of our graph-based approach: feature after the first stage of label propagation (Eq.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"RandLM\u2019s stupid backoff variant stores counts instead of probabilities and backoffs.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This leads to word- and constituent-boundaries discrepancy, which breaks the assumptions underlying current state-of-the-art statistical parsers.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The results of this experiment can be seen in Table 5.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"IRSTLM 5.60.02 (Federico et al., 2008) is a sorted trie implementation designed for lower memory consumption.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"More examples are shown in Figure 5.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We have shown the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This causes a problem for reverse trie implementations, including SRILM itself, because it leaves n+1-grams without an n-gram node pointing to them.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For example, ... fraud related to work on a federally funded sewage plant in Georgia In this case, Georgia is extracted: the NP containing it is a complement to the preposition in; the PP headed by in modifies the NP a federally funded sewage plant, whose head is the singular noun plant.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"For monolingual treebank data we relied on the CoNLL-X and CoNLL-2007 shared tasks on dependency parsing (Buchholz and Marsi, 2006; Nivre et al., 2007).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(Other classes handled by the current system are discussed in Section 5.)","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"With regard to purely morphological phenomena, certain processes are not han\u00c2\u00ad dled elegantly within the current framework Any process involving reduplication, for instance, does not lend itself to modeling by finite-state techniques, since there is no way that finite-state networks can directly implement the copying operations required.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Dropping the conditioning on 0 for brevity, and letting \u00afc\u03bb(s, t) = c\u03bb(s, t) + yu(s|t), and \u00afc\u03bb(t) = 4Note that the probabilities in (7) need only be evaluated over the support of \u02dcp(s, t), which is quite small when this distribution is derived from a dev set.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The first value reports use immediately after loading while the second reports the increase during scoring. dBerkeleyLM is written in Java which requires memory be specified in advance.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We adopted the MUC6 guidelines for evaluating coreference relationships based on transitivity in anaphoric chains.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Their results are then compared with the results of an automatic segmenter.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This token may further embed into a larger utterance, e.g., \u2018bcl hneim\u2019 (literally \u201cin-the-shadow the-pleasant\u201d, meaning roughly \u201cin the pleasant shadow\u201d) in which the dominated Noun is modified by a proceeding space-delimited adjective.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"First, we describe how the caseframes are represented and learned.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"This technique has the advantage of requiring no training, but it has the disadvantage of treating all parsers equally even though they may have differing accuracies or may specialize in modeling different phenomena.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"So far the graph has been completely unlabeled.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"basically complete, yet some improvements and extensions are still under way.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Table 3 contains the results for evaluating our systems on the test set (section 22).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Of course, we are primarily interested in applying our techniques to languages for which no labeled resources are available.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For example, out of 905 phrases in the CC- domain, 211 phrases contain keywords found in step 2.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The bias of automatic methods in favor of statistical systems seems to be less pronounced on out-of-domain test data.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The type of alignment we have considered so far requires the same length for source and target sentence, i.e. I = J. Evidently, this is an unrealistic assumption, therefore we extend the concept of inverted alignments as follows: When adding a new position to the coverage set C, we might generate either \u00c3\u0086 = 0 or \u00c3\u0086 = 1 new target words.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Dropping the conditioning on 0 for brevity, and letting \u00afc\u03bb(s, t) = c\u03bb(s, t) + yu(s|t), and \u00afc\u03bb(t) = 4Note that the probabilities in (7) need only be evaluated over the support of \u02dcp(s, t), which is quite small when this distribution is derived from a dev set.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Fort= 1,...,T:","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"This work has been supported as part of the Verbmobil project (contract number 01 IV 601 A) by the German Federal Ministry of Education, Science, Research and Technology and as part of the Eutrans project (ESPRIT project number 30268) by the European Community.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"AdaBoost is given access to a weak learning algorithm, which accepts as input the training examples, along with a distribution over the instances.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"(S; C; j); Not only the coverage set C and the positions j; j0, but also the verbgroup states S; S0 are taken into account.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Finally, as\u00c2\u00ad suming a simple bigram backoff model, we can derive the probability estimate for the particular unseen word i\u00c2\u00a51J1l.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In our situation, the competing hypotheses are the possible antecedents for an anaphor.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses information extraction patterns to identify contextual roles and creates four contextual role knowledge sources using unsupervised learning.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The additional morphological material in such cases appears after the stem and realizes the extended meaning.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"SSER: subjective sentence error rate: For a more detailed analysis, the translations are judged by a human test person.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"At each point during the derivation, the parser has a choice between pushing the next input token onto the stack \u2013 with or without adding an arc from the token on top of the stack to the token pushed \u2013 and popping a token from the stack \u2013 with or without adding an arc from the next input token to the token popped.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We then use linguistic and annotation insights to develop a manually annotated grammar for Arabic (\u00c2\u00a74).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The family name set is restricted: there are a few hundred single-hanzi family names, and about ten double-hanzi ones.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Vocabulary lookup is a sorted array of 64-bit word hashes.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"BABAR applies a DempsterShafer probabilistic model to make resolutions based on evidence from the contextual role knowledge sources as well as general knowledge sources.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But the city name Sharm Al- Sheikh is also iDafa, hence the possibility for the incorrect annotation in (b).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"We can do that . IbmS: Yes, wonderful.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"More recently, Subramanya et al. (2010) defined a graph over the cliques in an underlying structured prediction model.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For this example, since the sequence Even News Broadcasting Corp. only appears once in the document, its longest substring that occurs in the same document is News Broadcasting Corp. In this case, News has an additional feature of I begin set to 1, Broadcasting has an additional feature of I continue set to 1, and Corp. has an additional feature of I end set to 1.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"In this paper we have proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"60 ( P e tr o v, 2 0 0 9 ) all B e r k e l e y ( S e p . 0 9 ) B a s e l i n e 7 0 a l l G o l d P O S 70 \u00e2\u0080\u0094 \u00e2\u0080\u0094 \u00e2\u0080\u0094 0 . 8 0 9 0.839 335 0 . 7 9","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Future work along these lines will incorporate other layers of annotation, in particular the syntax information.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"They are also labelled for their topicality (yes \/ no), and this annotation is accompanied by a confidence value assigned by the annotator (since it is a more subjective matter).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Almost all annotators expressed their preference to move to a ranking-based evaluation in the future.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Ignoring the identity of the target language words e and e0, the possible partial hypothesis extensions due to the IBM restrictions are shown in Table 2.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The PROBING model can perform optimistic searches by jumping to any n-gram without needing state and without any additional memory.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Since these are distinct syntactic units, they are typically segmented.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In this domain the major scenarios involve the things they agreed on, rather than the mere fact that they agreed.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Ideally we\u00e2\u0080\u0099d like to know the thematic role of each extracted noun phrase, but AutoSlog does not generate thematic roles.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Note that the good classes JADE, GOLD and GRASS have lower costs than the bad classes SICKNESS, DEATH and RAT, as desired, so the trend observed for the results of this method is in the right direction.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The function A : F \u2014* C maps from the language specific fine-grained tagset F to the coarser universal tagset C and is described in detail in \u00a76.2: Note that when tx(y) = 1 the feature value is 0 and has no effect on the model, while its value is \u2212oc when tx(y) = 0 and constrains the HMM\u2019s state space.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For example, it is well-known that one can build a finite-state bigram (word) model by simply assigning a state Si to each word Wi in the vocabulary, and having (word) arcs leaving that state weighted such that for each Wj and corresponding arc aj leaving Si, the cost on aj is the bigram cost of WiWj- (Costs for unseen bigrams in such a scheme would typically be modeled with a special backoff state.)","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Fig.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"This result suggests that all of contextual role KSs can provide useful information for resolving anaphora.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"2.2.2 The Caseframe Network The first type of contextual role knowledge that BABAR learns is the Caseframe Network (CFNet), which identifies caseframes that co-occur in anaphor\/antecedent resolutions.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"In this way we restrict the parameterization of a Language Original case English Danish Dutch German Spanish Swedish Portuguese 94.6 96.3 96.6 95.5 95.4 93.3 95.6 Table 1: Upper bound on tagging accuracy assuming each word type is assigned to majority POS tag.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The algorithm builds two classifiers in parallel from labeled and unlabeled data.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Sheffield Guthrie and Hepple (2010) explore several randomized compression techniques, but did not release code.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We of course also fail to identify, by the methods just described, given names used without their associated family name.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In our work, we demonstrate that using a simple na\u00c2\u00a8\u00c4\u00b1veBayes approach also yields substantial performance gains, without the associated training complexity.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"To quantize, we use the binning method (Federico and Bertoldi, 2006) that sorts values, divides into equally sized bins, and averages within each bin.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Therefore, the number of fine tags varied across languages for our experiments; however, one could as well have fixed the set of HMM states to be a constant across languages, and created one mapping to the universal POS tagset.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Presence of the determiner J Al. 2.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Generalizing state minimization, the model could also provide explicit bounds on probability for both backward and forward extension.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Lexical-knowledge-based approaches that include statistical information generally presume that one starts with all possible segmentations of a sentence, and picks the best segmentation from the set of possible segmentations using a probabilistic or cost\u00c2\u00ad based scoring mechanism.","label":0},{"source_text":"Here we present two algorithms.","target_text":"It was motivated by the observation that the (Yarowsky 95) algorithm added a very large number of rules in the first few iterations.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This allow the learners to &quot;bootstrap&quot; each other by filling the labels of the instances on which the other side has abstained so far.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The results of this experiment can be seen in Table 5.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"4.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"This work was supported in part by the National Science Foundation under grant IRI9704240.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"25 16.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The paper is organized as follows: Section 2 explains the different layers of annotation that have been produced or are being produced.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"\u00e2\u0080\u00a2 Some tools would allow for the desired annotation mode, but are so complicated (they can be used for many other purposes as well) that annotators take a long time getting used to them.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Unlike GCFG's, however, the composition operations of LCFRS's are restricted to be linear (do not duplicate unboundedly large structures) and nonerasing (do not erase unbounded structures, a restriction made in most modern transformational grammars).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"However, we note that the final conditional estimates p(s|t) from a given phrase table maximize the likelihood of joint empirical phrase pair counts over a word-aligned corpus.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Because these two words have identical complements, syntax rules are typically unhelpful for distinguishing between them.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Our goal is to define a class of formal systems, and show that any member of this class will possess certain attractive properties.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The orthographic normalization strategy we use is simple.10 In addition to removing all diacritics, we strip instances of taTweel J=J4.i, collapse variants of alif to bare alif,11 and map Ara bic punctuation characters to their Latin equivalents.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This is an iterative method that improves the estimation of the parameters at each iteration.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The procedure using the tagged sentences to discover paraphrases takes about one hour on a 2GHz Pentium 4 PC with 1GB of memory.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"f, nan2gual+men0 'pumpkins' is by no means impossible.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"See Table 2 for the tag set size of other languages.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Across eight European languages, our approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"By design, they readily capture regularities at the token-level.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR then computes statistics over the training examples measuring the frequency with which extraction patterns and noun phrases co-occur in coreference resolutions.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","label":1},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Experiments in two domains showed that the contextual role knowledge improved coreference performance, especially on pronouns.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In order to create an IE system for a new domain, one has to spend a long time to create the knowledge.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As we have noted in Section 2, the general semantic class to which a hanzi belongs is often predictable from its semantic radical.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Morphological analyzers for Hebrew that analyze a surface form in isolation have been proposed by Segal (2000), Yona and Wintner (2005), and recently by the knowledge center for processing Hebrew (Itai et al., 2006).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"To combat the proliferation of parsing edges, we prune the lattices according to a hand-constructed lexicon of 31 clitics listed in the ATB annotation guidelines (Maamouri et al., 2009a).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"To initialize the graph for label propagation we use a supervised English tagger to label the English side of the bitext.7 We then simply count the individual labels of the English tokens and normalize the counts to produce tag distributions over English word types.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We extend the Matsoukas et al approach in several ways.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For the examples given in (1) and (2) this certainly seems possible.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"A promising direction for future work is to explicitly model a distribution over tags for each word type.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Note that hanzi that are not grouped into dictionary words (and are not identified as single\u00c2\u00ad hanzi words), or into one of the other categories of words discussed in this paper, are left unattached and tagged as unknown words.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Finally, model GTv = 2 includes parent annotation on top of the various state-splits, as is done also in (Tsarfaty and Sima\u2019an, 2007; Cohen and Smith, 2007).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Graph construction does not require any labeled data, but makes use of two similarity functions.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The focus of our work is on the use of contextual role knowledge for coreference resolution.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"As each global feature group is added to the list of features, we see improvements to both MUC6 and MUC6 MUC7 Baseline 90.75% 85.22% + ICOC 91.50% 86.24% + CSPP 92.89% 86.96% + ACRO 93.04% 86.99% + SOIC 93.25% 87.22% + UNIQ 93.27% 87.24% Table 3: F-measure after successive addition of each global feature group Table 5: Comparison of results for MUC6 Systems MUC6 MUC7 No.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Table 1 shows four words \u00e2\u0080\u009c 0 Indeed NN Indeed Saddamwhose unvocalized surface forms 0 an are indistinguishable.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Methods that allow multiple segmentations must provide criteria for choosing the best segmentation.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We checked whether the discovered links are listed in WordNet.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Table 9 shows that MADA produces a high quality segmentation, and that the effect of cascading segmentation errors on parsing is only 1.92% F1.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"As the name implies, space is O(m) and linear in the number of entries.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The average individual parser accuracy was reduced by more than 5% when we added this new parser, but the precision of the constituent voting technique was the only result that decreased significantly.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"La ng ua ge # To ke ns # W or d Ty pe s # Ta gs E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 1 1 7 3 7 6 6 9 4 3 8 6 2 0 3 5 6 8 6 9 9 6 0 5 2 0 6 6 7 8 8 9 3 3 4 1 9 1 4 6 7 4 9 2 0 6 1 8 3 5 6 2 8 3 9 3 7 2 3 2 5 2 8 9 3 1 1 6 4 5 8 2 0 0 5 7 4 5 2 5 1 2 5 4 2 2 4 7 4 1 Table 2: Statistics for various corpora utilized in experiments.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"It is worth noting that the middle words of the Italian trigrams are nouns too, which exhibits the fact that the similarity metric connects types having the same syntactic category.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The distinctions in the ATB are linguistically justified, but complicate parsing.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The first method uses a similar algorithm to that of (Yarowsky 95), with modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"f;g denotes the empty set, where no source sentence position is covered.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We are especially grateful to Taylor Berg- Kirkpatrick for running additional experiments.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Otherwise, the scope of the search problem shrinks recursively: if A[pivot] < k then this becomes the new lower bound: l +\u2014 pivot; if A[pivot] > k then u +\u2014 pivot.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"If the same pair of NE instances is used with different phrases, these phrases are likely to be paraphrases.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"ments contained 322 anaphoric links.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Several systems propose statistical methods for handling unknown words (Chang et al. 1992; Lin, Chiang, and Su 1993; Peng and Chang 1993).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Section 5 covers relevant previous work on SMT adaptation, and section 6 concludes.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"rently, some annotations (in particular the connectives and scopes) have already moved beyond the core corpus; the others will grow step by step.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The following algorithm was then used to induce new rules: Let Count' (x) be the number of times feature x is seen with some known label in the training data.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"As one can see, only the trigrams [suo incarceramento ,], [suo iter ,] and [suo carattere ,] are connected to English words.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We evaluate our model on seven languages exhibiting substantial syntactic variation.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"This approach is similar to BABAR in that they both acquire knowledge from earlier resolutions.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"It is not easy to make a clear definition of \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Note that while the standard HMM, has O(K n) emission parameters, our model has O(n) effective parameters.3 Token Component Once HMM parameters (\u00cf\u0086, \u00ce\u00b8) have been drawn, the HMM generates a token-level corpus w in the standard way: P (w, t|\u00cf\u0086, \u00ce\u00b8) = P (T , W , \u00ce\u00b8, \u00cf\u0088, \u00cf\u0086, t, w|\u00ce\u00b1, \u00ce\u00b2) = P (T , W , \u00cf\u0088|\u00ce\u00b2) [Lexicon] \u00ef\u00a3\u00ab n n \u00ef\u00a3\u00ad (w,t)\u00e2\u0088\u0088(w,t) j \u00ef\u00a3\u00b6 P (tj |\u00cf\u0086tj\u00e2\u0088\u00921 )P (wj |tj , \u00ce\u00b8tj )\u00ef\u00a3\u00b8 P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) [Parameter] P (w, t|\u00cf\u0086, \u00ce\u00b8) [Token] We refer to the components on the right hand side as the lexicon, parameter, and token component respectively.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For example, in the CC-domain, 96 keywords are found which have TF\/ITF scores above a threshold; some of them are shown in Figure 3.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We extend the Matsoukas et al approach in several ways.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Even this may be nondeterministic, in case the graph contains several non-projective arcs whose lifts interact, but we use the following algorithm to construct a minimal projective transformation D0 = (W, A0) of a (nonprojective) dependency graph D = (W, A): The function SMALLEST-NONP-ARC returns the non-projective arc with the shortest distance from head to dependent (breaking ties from left to right).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"In this case we are interested in finding' the maximum probability parse, ri, and Mi is the set of relevant (binary) parsing decisions made by parser i. ri is a parse selected from among the outputs of the individual parsers.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In the rhetorical tree, nuclearity information is then used to extract a \u00e2\u0080\u009ckernel tree\u00e2\u0080\u009d that supposedly represents the key information from which the summary can be generated (which in turn may involve co-reference information, as we want to avoid dangling pronouns in a summary).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We have not yet tried this.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The iw all map variant uses a non-0 y weight on a uniform prior in p,,(s t), and outperforms a version with y = 0 (iw all) and the \u201cflattened\u201d variant described in section 3.2.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"MENE (without the help of hand-coded systems) has been shown to be somewhat inferior in performance.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses unsupervised learning to acquire this knowledge from plain text without the need for annotated training data.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"5 60.6 Table 3: Multilingual Results: We report token-level one-to-one and many-to-one accuracy on a variety of languages under several experimental settings (Section 5).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The basic word order is VSO, but SVO, VOS, and VO configurations are also possible.2 Nouns and verbs are created by selecting a consonantal root (usually triliteral or quadriliteral), which bears the semantic core, and adding affixes and diacritics.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We realize the importance of paraphrase; however, the major obstacle is the construction of paraphrase knowledge.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Each treelet (an internal node with all its children) represents the use of a rule that is encapsulated by the grammar The grammar encapsulates (either explicitly or implicitly) a finite number of rules that can be written as follows: n > 0 In the case of CFG's, for each production In the case of TAG's, a derivation step in which the derived trees RI, \u2022 \u2022 \u2022 , On are adjoined into fi at rhe addresses \u2022 \u2022 \u2022 \u2022 in. would involve the use of the following rule2.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In a fully supervised setting, the task is to learn a function f such that for all i = 1...m, f (xi,i, 12,i) = yz.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We construct a mapping from all the space-delimited tokens seen in the training sentences to their corresponding analyses.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Mikheev et al.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"All commentaries have been annotated with rhetorical structure, using RSTTool4 and the definitions of discourse relations provided by Rhetorical Structure Theory (Mann, Thompson 1988).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Thus, we feel fairly confident that for the examples we have considered from Gan's study a solution can be incorporated, or at least approximated, within a finite-state framework.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"As is standard, we report the greedy one-to-one (Haghighi and Klein, 2006) and the many-to-one token-level accuracy obtained from mapping model states to gold POS tags.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"In addition to using the simple features directly, we also trained an SVM classifier with these features to distinguish between IN and OUT phrase pairs.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"If 0(A) gives the number of occurrences of each terminal in the structure named by A, then, given the constraints imposed on the formalism, for each rule A --. fp(Ai, , An) we have the equality where c\u201e is some constant.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We are currently exploring other methods that employ similar ideas and their formal properties.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Without using the same test corpus, direct comparison is obviously difficult; fortunately, Chang et al. include a list of about 60 sentence fragments that exemplify various categories of performance for their system.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Let H be the set of hanzi, p be the set of pinyin syllables with tone marks, and P be the set of grammatical part-of-speech labels.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This latter evaluation compares the performance of the system with that of several human judges since, as we shall show, even people do not agree on a single correct way to segment a text.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Phrase-level granularity distinguishes our work from previous work by Matsoukas et al (2009), who weight sentences according to sub-corpus and genre membership.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"17 They also provide a set of title-driven rules to identify names when they occur before titles such as $t. 1: xianlshengl 'Mr.' or i:l:itr!J tai2bei3 shi4zhang3 'Taipei Mayor.'","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"(Again, the goal of also in structural features.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"BerkeleyLM uses states to optimistically search for longer n-gram matches first and must perform twice as many random accesses to retrieve backoff information.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Both authors are members of the Center for Language and Speech Processing at Johns Hopkins University.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Here a set is represented by the keyword and the number in parentheses indicates the number of shared NE pair instances.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Every parse \u03c0 selects a specific morphological segmentation (l1...lk) (a path through the lattice).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The cities of the traveling salesman problem correspond to source Table 1: DP algorithm for statistical machine translation.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Experiments in two domains showed that the contextual role knowledge improved coreference performance, especially on pronouns.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Cluster phrases based on Links We now have a set of phrases which share a keyword.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"85 82.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In addition, as the CRF and PCFG look at similar sorts of information from within two inherently different models, they are far from independent and optimizing their product is meaningless.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"An edge indicates that the two features must have the same label.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Corpus Step 1 NE pair instances Step 2 Step 1.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Chinese word segmentation can be viewed as a stochastic transduction problem.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"There has been a lot of research on such lexical relations, along with the creation of resources such as WordNet.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"This assumption, however, is not inherent to type-based tagging models.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Therefore, we want state to encode the minimum amount of information necessary to properly compute language model scores, so that the decoder will be faster and make fewer search errors.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"On each language we investigate the contribution of each component of our model.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Out of those 15 links, 4 are errors, namely \u00e2\u0080\u009cbuy - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cpurchase - stake\u00e2\u0080\u009d \u00e2\u0080\u009cacquisition - stake\u00e2\u0080\u009d.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"This fact annoyed especially his dog...).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Two sets of examples from Gan are given in (1) and (2) (:::::: Gan's Appendix B, exx.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Consequently, we implemented our own annotation tool ConAno in Java (Stede, Heintze 2004), which provides specifically the functionality needed for our purpose.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Like verbs, maSdar takes arguments and assigns case to its objects, whereas it also demonstrates nominal characteristics by, e.g., taking determiners and heading iDafa (Fassi Fehri, 1993).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"These two restrictions impose the constraint that the result of composing any two structures should be a structure whose &quot;size&quot; is the sum of its constituents plus some constant For example, the operation 4, discussed in the case of CFG's (in Section 4.1) adds the constant equal to the sum of the length of the strings VI, un+r\u2022 Since we are considering formalisms with arbitrary structures it is difficult to precisely specify all of the restrictions on the composition operations that we believe would appropriately generalize the concatenation operation for the particular structures used by the formalism.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The focus of our work is on the use of contextual role knowledge for coreference resolution.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We show that noun-noun vs. discourse-level coordination ambiguity in Arabic is a significant source of parsing errors (Table 8c).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The idea is to have a pipeline of shallow-analysis modules (tagging, chunk- ing, discourse parsing based on connectives) and map the resulting underspecified rhetorical tree (see Section 2.4) into a knowledge base that may contain domain and world knowledge for enriching the representation, e.g., to resolve references that cannot be handled by shallow methods, or to hypothesize coherence relations.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We allow any number of bits from 2 to 25, unlike IRSTLM (8 bits) and BerkeleyLM (17\u221220 bits).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"At this stage the lattice path corresponds to segments only, with no PoS assigned to them.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Corporate Suffixes and Person Prefixes of Other Occurrences (CSPP): If McCann has been seen as Mr. McCann somewhere else in the document, then one would like to give person a higher probability than organization.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"For instance, by altering the emission distribution parameters, Johnson (2007) encourages the model to put most of the probability mass on few tags.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For example, it is well-known that one can build a finite-state bigram (word) model by simply assigning a state Si to each word Wi in the vocabulary, and having (word) arcs leaving that state weighted such that for each Wj and corresponding arc aj leaving Si, the cost on aj is the bigram cost of WiWj- (Costs for unseen bigrams in such a scheme would typically be modeled with a special backoff state.)","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For example, in Northern dialects (such as Beijing), a full tone (1, 2, 3, or 4) is changed to a neutral tone (0) in the final syllable of many words: Jll donglgual 'winter melon' is often pronounced donglguaO.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The idea of distinguishing between general and domain-specific examples is due to Daum\u00b4e and Marcu (2006), who used a maximum-entropy model with latent variables to capture the degree of specificity.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Evaluating parsing results in our joint framework, as argued by Tsarfaty (2006), is not trivial under the joint disambiguation task, as the hypothesized yield need not coincide with the correct one.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Similar behavior is observed when adding features.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Various segmentation approaches were then compared with human performance: 1.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"shows some keywords with their scores.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"from the subset of the United Informatics corpus not used in the training of the models.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Gather phrases using keywords Next, we select a keyword for each phrase \u00e2\u0080\u0093 the top-ranked word based on the TF\/IDF metric.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Chunking is not enough to find such relationships.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Tag set As is standard, for all experiments, we set the number of latent model tag states to the size of the annotated tag set.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Pseudo-Projective Dependency Parsing","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Since the inclusion of out-ofdomain test data was a very late decision, the participants were not informed of this.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For example, in Figure 3, we can see that the phrases in the \u00e2\u0080\u009cbuy\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire\u00e2\u0080\u009d and \u00e2\u0080\u009cpurchase\u00e2\u0080\u009d sets are mostly paraphrases.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Furthermore, the systematic way in which particles are prefixed to one another and onto an open-class category gives rise to a distinct sort of morphological ambiguity: space-delimited tokens may be ambiguous between several different segmentation possibilities.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The best analysis of the corpus is taken to be the true analysis, the frequencies are re-estimated, and the algorithm is repeated until it converges.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"18 We are grateful to ChaoHuang Chang for providing us with this set.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Our results suggest that current parsing models would benefit from better annotation consistency and enriched annotation in certain syntactic configurations.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Syntactic universals are a well studied concept in linguistics (Carnie, 2002; Newmeyer, 2005), and were recently used in similar form by Naseem et al. (2010) for multilingual grammar induction.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Second, BABAR performs reliable case resolution to identify anaphora that can be easily resolved using the lexical and syntactic heuristics described in Section 2.1.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Dagan and Itai (Dagan and Itai, 1990) experimented with co-occurrence statistics that are similar to our lexical caseframe expectations.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In fact, during the first rounds many of the predictions of Th., g2 are zero.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In addition we report for each model its performance on goldsegmented input (GS) to indicate the upper bound 11Overt definiteness errors may be seen as a wrong feature rather than as wrong constituent and it is by now an accepted standard to report accuracy with and without such errors. for the grammars\u2019 performance on the parsing task.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For example, in the sentence that starts with \u00e2\u0080\u009cBush put a freeze on . . .","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Finally, we note that simple weighting gives nearly a 2% F1 improvement, whereas Goldberg and Tsarfaty (2008) found that unweighted lattices were more effective for Hebrew.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"On the other hand, given that all schemes have similar parsing accuracy overall, this means that the Path scheme is the least likely to introduce errors on projective arcs.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Note that these observa sider suffix features, capitalization features, punctuation, and digit features.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"TAG's can be shown to be equivalent to this restricted system.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"They are also labelled for their topicality (yes \/ no), and this annotation is accompanied by a confidence value assigned by the annotator (since it is a more subjective matter).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Our representation of contextual roles is based on information extraction patterns that are converted into simple caseframes.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Parent Head Modif er Dir # gold F1 Label # gold F1 NP NP TAG R 946 0.54 ADJP 1216 59.45 S S S R 708 0.57 SBAR 2918 69.81 NP NP ADJ P R 803 0.64 FRAG 254 72.87 NP NP N P R 2907 0.66 VP 5507 78.83 NP NP SBA R R 1035 0.67 S 6579 78.91 NP NP P P R 2713 0.67 PP 7516 80.93 VP TAG P P R 3230 0.80 NP 34025 84.95 NP NP TAG L 805 0.85 ADVP 1093 90.64 VP TAG SBA R R 772 0.86 WHN P 787 96.00 S VP N P L 961 0.87 (a) Major phrasal categories (b) Major POS categories (c) Ten lowest scoring (Collins, 2003)-style dependencies occurring more than 700 times Table 8: Per category performance of the Berkeley parser on sentence lengths \u00e2\u0089\u00a4 70 (dev set, gold segmentation).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The next two rows are results of oracle experiments.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"needs to be in initCaps to be considered for this feature.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Before presenting our results, we describe the datasets that we used, as well as two baselines.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For each resolution in the training data, BABAR also associates the co-referring expression of an NP with the NP\u00e2\u0080\u0099s caseframe.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Figure 2 shows a screenshot (which is of somewhat limited value, though, as color plays a major role in signalling the different statuses of the information).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Recently, lattices have been used successfully in the parsing of Hebrew (Tsarfaty, 2006; Cohen and Smith, 2007), a Semitic language with similar properties to Arabic.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The Recency KS computes the distance between the candidate and the anaphor relative to its scope.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The CFLex and CFNet knowledge sources provide positive evidence that a candidate NP and anaphor might be coreferent.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"For example, we can easily imagine that the number of paraphrases for \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d is enormous and it is not possible to create a comprehensive inventory by hand.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In order to create an IE system for a new domain, one has to spend a long time to create the knowledge.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(For some recent corpus-based work on Chinese abbreviations, see Huang, Ahrens, and Chen [1993].)","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The performance was 80.99% recall and 61.83% precision.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The theory has also been validated empirically.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Morphological processes in Semitic languages deliver space-delimited words which introduce multiple, distinct, syntactic units into the structure of the input sentence.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Motivated by these questions, we significantly raise baselines for three existing parsing models through better grammar engineering.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"So we decided to use semantic class information only to rule out candidates.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Unexpectedly, because the ministries of treasury and education both had prepared the teacher plan together.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"It is formally straightforward to extend the grammar to include these names, though it does increase the likelihood of overgeneration and we are unaware of any working systems that incorporate this type of name.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The following auxiliary quantity is defined: Qe0 (e; C; j) := probability of the best partial hypothesis (ei 1; bi 1), where C = fbkjk = 1; ; ig, bi = j, ei = e and ei\u00f4\u0080\u0080\u00801 = e0.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The PROBING model is 2.4 times as fast as the fastest alternative, SRILM, and uses less memory too.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"It can be seen that M performs a top-down recognition of the input al ... nin logspace.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"This annotation choice weakens splitIN.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The most informative scheme, Head+Path, gives the highest scores, although with respect to Head the difference is not statistically significant, while the least informative scheme, Path \u2013 with almost the same performance on treebank transformation \u2013 is significantly lower (p < 0.01).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"To resolve the anaphor, we survey the final belief values assigned to each candidate\u00e2\u0080\u0099s singleton set.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Each set is assigned two values: belief and plausibility.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"vierten 12.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"For all these annotation taks, G\u00c2\u00a8otze developed a series of questions (essentially a decision tree) designed to lead the annotator to the ap propriate judgement.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Except our own and MENE + reference resolution, the results in Table 6 are all official MUC7 results.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Global context from the whole document is available and can be exploited in a natural manner with a maximum entropy classifier.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"We believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The web-based Annis imports data in a variety of XML formats and tagsets and displays it in a tier-orientedway (optionally, trees can be drawn more ele gantly in a separate window).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Section 5 covers relevant previous work on SMT adaptation, and section 6 concludes.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"\u00ce\u00b2 is the shared hyperparameter for the tag assignment prior and word feature multinomials.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The 4th block contains instance-weighting models trained on all features, used within a MAP TM combination, and with a linear LM mixture.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The alignment model uses two kinds of parameters: alignment probabilities p(aj jaj\u00f4\u0080\u0080\u00801; I; J), where the probability of alignment aj for position j depends on the previous alignment position aj\u00f4\u0080\u0080\u00801 (Ney et al., 2000) and lexicon probabilities p(fj jeaj ).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For the purposes of EM, the &quot;observed&quot; data is {(xi, Ya\u2022 \u2022 \u2022 (xrn, Yrn), xfil, and the hidden data is {ym+i y}.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Section 4 presents experimen tal results on two corpora: the MUC4 terrorism corpus, and Reuters texts about natural disasters.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The extent to which this constraint is enforced varies greatly across existing methods.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Because conjunctions are elevated in the parse trees when they separate recursive constituents, we choose the right sister instead of the category of the next word.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The equation for sampling a single type-level assignment Ti is given by, 0.2 0 5 10 15 20 25 30 Iteration Figure 2: Graph of the one-to-one accuracy of our full model (+FEATS) under the best hyperparameter setting by iteration (see Section 5).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"1 2 3.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Sie.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"All three models evaluated in this paper incorrectly analyze the constituent as iDafa; none of the models attach the attributive adjectives properly.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The 14 general-language features embody straightforward cues: frequency, \u201ccentrality\u201d as reflected in model scores, and lack of burstiness.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The distribution specifies the relative weight, or importance, of each example \u2014 typically, the weak learner will attempt to minimize the weighted error on the training set, where the distribution specifies the weights.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We elected run Moses single-threaded to minimize the impact of RandLM\u2019s cache on memory use.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Comparison with state-of-the-art taggers For comparison we consider two unsupervised tag- gers: the HMM with log-linear features of Berg- Kirkpatrick et al.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In speech recognition the arcs of the lattice are typically weighted in order to indicate the probability of specific transitions.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In our experiment, we set the threshold of the TF\/ITF score empirically using a small development corpus; a finer adjustment of the threshold could reduce the number of such keywords.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"88,962 (spelling,context) pairs were extracted as training data.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Some relations are signalled by subordinating conjunctions, which clearly demarcate the range of the text spans related (matrix clause, embedded clause).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Base NPs are the other significant category of nominal phrases.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"input: source string f1:::fj :::fJ initialization for each cardinality c = 1; 2; ; J do for each pair (C; j), where j 2 C and jCj = c do for each target word e 2 E Qe0 (e; C; j) = p(fj je) max \u00c3\u0086;e00 j02Cnfjg fp(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j0)g words fj in the input string of length J. For the final translation each source position is considered exactly once.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Ex: He was found in San Jose, where ...","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The development set contained 44088 constituents in 2416 sentences and the test set contained 30691 constituents in 1699 sentences.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We can only compare with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Figure 1 reveals that an event that \u00e2\u0080\u009cdamaged\u00e2\u0080\u009d objects may also cause injuries; a disaster that \u00e2\u0080\u009coccurred\u00e2\u0080\u009d may be investigated to find its \u00e2\u0080\u009ccause\u00e2\u0080\u009d; a disaster may \u00e2\u0080\u009cwreak\u00e2\u0080\u009d havoc as it \u00e2\u0080\u009ccrosses\u00e2\u0080\u009d geographic regions; and vehicles that have a \u00e2\u0080\u009cdriver\u00e2\u0080\u009d may also \u00e2\u0080\u009ccarry\u00e2\u0080\u009d items.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each domain, we created a semantic dictionary by doing two things.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The 8 similarity-to-IN features are based on word frequencies and scores from various models trained on the IN corpus: To avoid numerical problems, each feature was normalized by subtracting its mean and dividing by its standard deviation.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The resulting algorithm is depicted in Table 1.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For example, as Gan (1994) has noted, one can construct examples where the segmen\u00c2\u00ad tation is locally ambiguous but can be determined on the basis of sentential or even discourse context.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Furthermore we expect the label distributions on the foreign to be fairly noisy, because the graph constraints have not been taken into account yet.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"pronunciations of individual words; they also need to compute intonational phrase boundaries in long utterances and assign relative prominence to words in those utterances.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We do not show the numbers for the Bayes models in Table 2 because the parameters involved were established using this set.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"First, any grammar must involve a finite number of elementary structures, composed using a finite number of composition operations.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Recently, combination techniques have been investigated for part of speech tagging with positive results (van Halteren et al., 1998; Brill and Wu, 1998).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Our first algorithm is similar to Yarowsky's, but with some important modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We quantify error categories in both evaluation settings.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"From here on we will refer to the named-entity string itself as the spelling of the entity, and the contextual predicate as the context.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Nonstochastic lexical-knowledge-based approaches have been much more numer\u00c2\u00ad ous.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Nonetheless there is no alternative to Reiche\u00e2\u0080\u0099s plan.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Among the IS-units, the referring expressions are marked as such and will in the second phase receive a label for cognitive status (active, accessible- text, accessible-situation, inferrable, inactive).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Our results outperform strong unsupervised baselines as well as approaches that rely on direct projections, and bridge the gap between purely supervised and unsupervised POS tagging models.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"This limitation is the obstacle to making the technology \u00e2\u0080\u009copen domain\u00e2\u0080\u009d.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We weight edges using a unigram language model estimated with Good- Turing smoothing.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Fortunately, we were able to obtain a copy of the full set of sentences from Chang et al. on which Wang, Li, and Chang tested their system, along with the output of their system.18 In what follows we will discuss all cases from this set where our performance on names differs from that of Wang, Li, and Chang.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Even with vocalization, there are linguistic categories that are difficult to identify without semantic clues.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"For simplicity, we assume that OUT is homogeneous.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"ICOC and CSPP contributed the greatest im provements.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The effect of the pruning threshold t0 is shown in Table 5.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, they list two sets, one consisting of 28 fragments and the other of 22 fragments, in which they had 0% recall and precision.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"At present, the \u00e2\u0080\u0098Potsdam Commentary Corpus\u00e2\u0080\u0099 (henceforth \u00e2\u0080\u0098PCC\u00e2\u0080\u0099 for short) consists of 170 commentaries from Ma\u00c2\u00a8rkische Allgemeine Zeitung, a German regional daily.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Training under this model involves estimation of parameter values for P(y), P(m) and P(x I y).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"And indeed, converging on annotation guidelines is even more difficult than it is with co-reference.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Therefore, a populated probing hash table consists of an array of buckets that contain either one entry or are empty.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Sc opi ng filters candidate if outside the anaphor\u00e2\u0080\u0099s scope.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"However, reads in the TRIE data structure are more expensive due to bit-level packing, so we found that it is faster to use interpolation search the entire time.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Because b is a function, no additional hypothesis splitting happens.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Statistical NERs usually find the sequence of tags that maximizes the probability , where is the sequence of words in a sentence, and is the sequence of named-entity tags assigned to the words in . Attempts have been made to use global information (e.g., the same named entity occurring in different sentences of the same document), but they usually consist of incorporating an additional classifier, which tries to correct the errors in the output of a first NER (Mikheev et al., 1998; Borthwick, 1999).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"9 50.2 +P RI OR be st me dia n 47.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Formally, the DempsterShafer theory defines a probability density function m(S), where S is a set of hypotheses.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The SynRole KS computes the relative frequency with which the candidates\u00e2\u0080\u0099 syntactic role (subject, direct object, PP object) appeared in resolutions in the training set.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"splitPUNC restores the convention of the WSJ.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"However, since we want to preserve as much of the original structure as possible, we are interested in finding a transformation that involves a minimal number of lifts.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Particular instances of relations are associated with goodness scores.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The sources of our dictionaries are listed in Table 2.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Yarowsky-cautious does not separate the spelling and contextual features, but does have a limit on the number of rules added at each stage.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Recall.","label":0},{"source_text":"Here we present two algorithms.","target_text":"2 We now introduce a new algorithm for learning from unlabeled examples, which we will call DLCoTrain (DL stands for decision list, the term Cotrain is taken from (Blum and Mitchell 98)).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Models To assess the marginal utility of each component of the model (see Section 3), we incremen- tally increase its sophistication.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"MENE has only been tested on MUC7.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"0 . 8 3 1 0.859 496 76.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For eight judges, ranging k between 1 and 8 corresponded to a precision score range of 90% to 30%, meaning that there were relatively few words (30% of those found by the automatic segmenter) on which all judges agreed, whereas most of the words found by the segmenter were such that one human judge agreed.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Finally, we find links between sets of phrases, based on the NE instance pair data (for example, different phrases which link \u00e2\u0080\u009cIBM\u00e2\u0080\u009d and \u00e2\u0080\u009cLotus\u00e2\u0080\u009d) (Step 4).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":".","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"One hopes that such a corpus will be forth\u00c2\u00ad coming.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"9 61.0 44.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Lexical rules are estimated in a similar manner.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The German finite verbs 'bin' (second example) and 'konnten' (third example) are too far away from the personal pronouns 'ich' and 'Sie' (6 respectively 5 source sentence positions).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"While possible to utilize the feature-based log-linear approach described in Berg-Kirkpatrick et al.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"State is implemented in their scrolling variant, which is a trie annotated with forward and backward pointers.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The pre terminal morphological analyses are mapped to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags provided with the tree- bank.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The performance was 80.99% recall and 61.83% precision.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Evaluation within a set The evaluation of paraphrases within a set of phrases which share a keyword is illustrated in Figure 4.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":", Sun day, then the feature DayOfTheWeek is set to 1.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We propose an unsupervised method to discover paraphrases from a large untagged corpus, without requiring any seed phrase or other cue.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We follow the guidelines developed in the TIGER project (Brants et al. 2002) for syntactic annotation of German newspaper text, using the Annotate3 tool for interactive construction of tree structures.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"This algorithm can be applied to statistical machine translation.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Since the lattice L for a given sentence W is determined by the morphological analyzer M we have which is precisely the formula corresponding to the so-called lattice parsing familiar from speech recognition.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"However, their system is a hybrid of hand-coded rules and machine learning methods.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Whether a language even has orthographic words is largely dependent on the writing system used to represent the language (rather than the language itself); the notion \"orthographic word\" is not universal.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Note that Wang, Li, and Chang's.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Without using the same test corpus, direct comparison is obviously difficult; fortunately, Chang et al. include a list of about 60 sentence fragments that exemplify various categories of performance for their system.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"A straightforward way to find the shortest tour is by trying all possible permutations of the n cities.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We asked six native speakers-three from Taiwan (TlT3), and three from the Mainland (M1M3)-to segment the corpus.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The secondary classifier in (Borthwick, 1999) uses information not just from the current article, but also from the whole test corpus, with an additional feature that indicates if the information comes from the same document or from another document.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"In our model, we associate these features at the type-level in the lexicon.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Many statistical or machine-learning approaches for natural language problems require a relatively large amount of supervision, in the form of labeled training examples.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The Verbmobil task is an appointment scheduling task.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"N, portion of examples on which both classifiers give a label rather than abstaining), and the proportion of these examples on which the two classifiers agree.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"These results are achieved by training on the official MUC6 and MUC7 training data, which is much less training data than is used by other machine learning systems that worked on the MUC6 or MUC7 named entity task (Bikel et al., 1997; Bikel et al., 1999; Borth- wick, 1999).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"6One of our experimental settings lacks document boundaries, and we used this approximation in both settings for consistency.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In contrast, our morphological probabilities are based on a unigram, lexeme-based model, and all other (local and non-local) contextual considerations are delegated to the PCFG.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Text generation, or at least the two phases of text planning and sentence planning, is a process driven partly by well-motivated choices (e.g., use this lexeme X rather than that more colloquial near-synonym Y ) and partly by con tation like that of PCC can be exploited to look for correlations in particular between syntactic structure, choice of referring expressions, and sentence-internal information structure.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"(If the TF\/IDF score of that word is below a threshold, the phrase is discarded.)","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"A graphical depiction of our model as well as a summary of random variables and parameters can be found in Figure 1.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We attain these results using several optimizations: hashing, custom lookup tables, bit-level packing, and state for left-to-right query patterns.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"While possible to utilize the feature-based log-linear approach described in Berg-Kirkpatrick et al.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"By definition, each existential NP uniquely specifies an object or concept, so we can infer that all instances of the same existential NP are coreferent (e.g., \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d always refers to the same entity).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"This limits the number of NE category pairs to 2,000 and the number of NE pair instances to 0.63 million.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The cities of the traveling salesman problem correspond to source Table 1: DP algorithm for statistical machine translation.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Even if an example like this is not labeled, it can be interpreted as a &quot;hint&quot; that Mr and president imply the same category.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"to explore how well we can induce POS tags using only the one-tag-per-word constraint.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Second, comparisons of different methods are not meaningful unless one can eval\u00c2\u00ad uate them on the same corpus.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"3) A tight coupling with the speech recognizer output.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Each pattern represents the role that a noun phrase plays in the surrounding context.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The method being described-henceforth ST..","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"For robustness reasons, the parser may output a set of dependency trees instead of a single tree. most dependent of the next input token, dependency type features are limited to tokens on the stack.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Our second contribution is to apply instance weighting at the level of phrase pairs.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The resulting algorithm has a complexity of O(n!).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"2.3 Assigning Evidence Values.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Figure 1 shows an example.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Thus it is possible, for illustration, to look for a noun phrase (syntax tier) marked as topic (information structure tier) that is in a bridging relation (co-reference tier) to some other noun phrase.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Inspection of the data shows that at n = 2500, the two classifiers both give labels on 44,281 (49.2%) of the unlabeled examples, and give the same label on 99.25% of these cases.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"J:j:l :zhongl :0.0 ;m,Jlong4 :0.0 (mHHaryg9tltHBI) \u00c2\u00a3: _ADV: 5.98 \u00c2\u00a5 :hua2:o.o E :_NC: 4.41 :mln2:o.o mm : guo2 : 0.0 (RopubllcofChlna) .....,.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"+ cost(unseen(fm, as desired.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"One hopes that such a corpus will be forth\u00c2\u00ad coming.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"able at http:\/\/nlp.stanford.edu\/projects\/arabic.shtml.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For sorted lookup, we compare interpolation search, standard C++ binary search, and standard C++ set based on red-black trees.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Te rro ris m Na tur al Dis ast ers mu rde r of < NP > kill ed <p atie nt > <a ge nt > da ma ged wa s inj ure d in < NP > <a ge nt > rep ort ed <a ge nt > add ed <a ge nt > occ urr ed cau se of < NP > <a ge nt > stat ed <a ge nt > add ed <a ge nt > wr eak ed <a ge nt > cro sse d per pet rat ed <p atie nt > con de mn ed <p atie nt > dri ver of < NP > <a ge nt > car ryi ng Figure 1: Caseframe Network Examples Figure 1 shows examples of caseframes that co-occur in resolutions, both in the terrorism and natural disaster domains.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Figure 4 shows the seven general knowledge sources (KSs) that represent features commonly used for coreference resolution.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We can now compare this algorithm to that of (Yarowsky 95).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In order to handle the necessary word reordering as an optimization problem within our dynamic programming approach, we describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming (Held, Karp, 1962).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For derived words that occur in our corpus we can estimate these costs as we would the costs for an underived dictionary entry.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We picked two domains, the CC-domain and the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain (PC-domain), for the evaluation, as the entire system output was too large to evaluate.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Our derivation is slightly different from the one presented in (Schapire and Singer 98) as we restrict at to be positive.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"So far the graph has been completely unlabeled.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Much confusion has been sown about Chinese writing by the use of the term ideograph, suggesting that hanzi somehow directly represent ideas.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Supervised methods have been applied quite successfully to the full MUC named-entity task (Bikel et al. 97).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Each of the constituents must have received at least 1 votes from the k parsers, so a > I1 and 2 \u2014 2k\u00b1-1 b > ri-5-111.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"termined by the category of the word that follows it.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Our method does not assume any knowledge about the target language (in particular no tagging dictionary is assumed), making it applicable to a wide array of resource-poor languages.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(student+plural) 'students,' which is derived by the affixation of the plural affix f, menD to the nounxue2shengl.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Along with IRSTLM and TPT, our binary format is memory mapped, meaning the file and in-memory representation are the same.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Following Dickinson (2005), we randomly sampled 100 variation nuclei from each corpus and evaluated each sample for the presence of an annotation error.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For terrorism, BABAR generated 5,078 resolutions: 2,386 from lexical seeding and 2,692 from syntactic seeding.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Statistics are printed before Moses exits and after parts of the decoder have been destroyed.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For each token , zero, one, or more of the features in each feature group are set to 1.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This leads to a linear combination of domain-specific probabilities, with weights in [0, 1], normalized to sum to 1.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"We are given a source string fJ 1 = f1:::fj :::fJ of length J, which is to be translated into a target string eI 1 = e1:::ei:::eI of length I. Among all possible target strings, we will choose the string with the highest probability: ^eI 1 = arg max eI 1 fPr(eI 1jfJ 1 )g = arg max eI 1 fPr(eI 1) Pr(fJ 1 jeI 1)g : (1) The argmax operation denotes the search problem, i.e. the generation of the output sentence in the target language.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"See Table 2 for the tag set size of other languages.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In considering recognition of LCFRS's, we make further assumption concerning the contribution of each structure to the input string, and how the composition operations combine structures in this respect.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The taggers were trained on datasets labeled with the universal tags.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"We also see that the increase in the size of the label sets for Head and Head+Path is far below the theoretical upper bounds given in Table 1.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(Blum and Mitchell 98) offer a promising formulation of redundancy, also prove some results about how the use of unlabeled examples can help classification, and suggest an objective function when training with unlabeled examples.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Our motivation for using DempsterShafer is that it provides a well-principled framework for combining evidence from multiple sources with respect to competing hypotheses.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Finally, we note that simple weighting gives nearly a 2% F1 improvement, whereas Goldberg and Tsarfaty (2008) found that unweighted lattices were more effective for Hebrew.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Thus we have some confidence that our own performance is at least as good as that of Chang et al.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Mixing, smoothing, and instance-feature weights are learned at the same time using an efficient maximum-likelihood procedure that relies on only a small in-domain development corpus.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This suggests a direct parallel to (1): where \u02dcp(s, t) is a joint empirical distribution extracted from the IN dev set using the standard procedure.2 An alternative form of linear combination is a maximum a posteriori (MAP) combination (Bacchiani et al., 2004).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"4 53.7 43.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The price shocked her.), or same-kind (e.g., Her health insurance paid for the hospital fees, but the automobile insurance did not cover the repair.).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"None of the parsers produce parses with crossing brackets, so none of them votes for both of the assumed constituents.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(Berland and Charniak 99) describe a method for extracting parts of objects from wholes (e.g., &quot;speedometer&quot; from &quot;car&quot;) from a large corpus using hand-crafted patterns.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"E.g. when 'Zahnarzttermin' is aligned to dentist's, the extended lexicon model might learn that 'Zahnarzttermin' actuallyhas to be aligned to both dentist's and ap pointment.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The ATB gives several different analyses to these words to indicate different types of coordination.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Figure 5 shows how this model is implemented as part of the dictionary WFST.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The most popular approach to dealing with seg\u00c2\u00ad mentation ambiguities is the maximum matching method, possibly augmented with further heuristics.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"To set \u03b2, we used the same criterion as for \u03b1, over a dev corpus: The MAP combination was used for TM probabilities only, in part due to a technical difficulty in formulating coherent counts when using standard LM smoothing techniques (Kneser and Ney, 1995).3 Motivated by information retrieval, a number of approaches choose \u201crelevant\u201d sentence pairs from OUT by matching individual source sentences from IN (Hildebrand et al., 2005; L\u00a8u et al., 2007), or individual target hypotheses (Zhao et al., 2004).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Tables 4 and 5 also show that putting all of the contextual role KSs in play at the same time produces the greatest performance gain.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Figure 4 shows the seven general knowledge sources (KSs) that represent features commonly used for coreference resolution.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"This feature ft incorporates information from the smoothed graph and prunes hidden states that are inconsistent with the thresholded vector tx.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We first give a brief overview of boosting algorithms.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The resulting parsers surpass the best previously published performance results for the Penn Treebank.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The adjunction operation with respect to tree sets (multicomponent adjunction) is defined as follows.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The distribution specifies the relative weight, or importance, of each example \u2014 typically, the weak learner will attempt to minimize the weighted error on the training set, where the distribution specifies the weights.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The work of Thatcher (1973) and Rounds (1969) define formal systems that generate tree sets that are related to CFG's and IG's.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"All improvements over the baseline are statistically significant beyond the 0.01 level (McNemar\u2019s test).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"An interesting observation is that while vertical markovization benefits all our models, its effect is less evident in Cohen and Smith.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The search starts in the hypothesis (I; f;g; 0).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The final strong hypothesis, denoted 1(x), is then the sign of a weighted sum of the weak hypotheses, 1(x) = sign (Vii atht(x)), where the weights at are determined during the run of the algorithm, as we describe below.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The work reported here is closely related to [Ha- segawa et al. 04].","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Based on these high-confidence alignments we can extract tuples of the form [u H v], where u is a foreign trigram type, whose middle word aligns to an English word type v. Our bilingual similarity function then sets the edge weights in proportion to these tuple counts.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The implementation is In a similar effort, (G\u00c2\u00a8otze 2003) developed a proposal for the theory-neutral annotation of information structure (IS) \u00e2\u0080\u0094 a notoriously difficult area with plenty of conflicting and overlapping terminological conceptions.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The orthographic normalization strategy we use is simple.10 In addition to removing all diacritics, we strip instances of taTweel J=J4.i, collapse variants of alif to bare alif,11 and map Ara bic punctuation characters to their Latin equivalents.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"F-measure is the harmonic mean of precision and recall, 2PR\/(P + R).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"If is unique, then a feature (Unique, Zone) is set to 1, where Zone is the document zone where appears.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Contains digits.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Finally, the focus\/background partition is annotated, together with the focus question that elicits the corresponding answer.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We can only compare with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"where the husband's family name is optionally prepended to the woman's full name; thus ;f:*lf#i xu3lin2-yan2hai3 would represent the name that Ms. Lin Yanhai would take if she married someone named Xu.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Table 9: Dev set results for sentences of length \u00e2\u0089\u00a4 70.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For definite NPs, the results are a mixed bag: some knowledge sources increased recall a little, but at the expense of some precision.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The natural baseline (baseline) outperforms the pure IN system only for EMEA\/EP fren.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Instead, we extend the variation n-gram method of Dickinson (2005) to compare annotation error rates in the WSJ and ATB.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We empirically control for the effect of our heuristics to make sure our pruning does not undermine the objectives of our joint task.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"As a partial solution, for pairs of hanzi that co-occur sufficiently often in our namelists, we use the estimated bigram cost, rather than the independence-based cost.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, the passive voice pattern \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d and the active voice pattern \u00e2\u0080\u009ckidnapped <direct object>\u00e2\u0080\u009d are merged into a single normalized pattern \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d.2 For the sake of sim plicity, we will refer to these normalized extraction patterns as caseframes.3 These caseframes can capture two types of contextual role information: (1) thematic roles corresponding to events (e.g, \u00e2\u0080\u009c<agent> kidnapped\u00e2\u0080\u009d or \u00e2\u0080\u009ckidnapped <patient>\u00e2\u0080\u009d), and (2) predicate-argument relations associated with both verbs and nouns (e.g., \u00e2\u0080\u009ckidnapped for <np>\u00e2\u0080\u009d or \u00e2\u0080\u009cvehicle with <np>\u00e2\u0080\u009d).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"2.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Memory usage is the same as with binary search and lower than with set.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For each domain, phrases which contain the same keyword are gathered to build a set of phrases (Step 3).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The first m pairs have labels yi, whereas for i = m + 1, , n the pairs are unlabeled.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(2) was extended to have an additional, innermost loop over the (3) possible labels.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Compared to the EMEA\/EP setting, the two domains in the NIST setting are less homogeneous and more similar to each other; there is also considerably more IN text available.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For example, in the phrase \u00e2\u0080\u009cCompany-A last week purchased rival Marshalls from Company-B\u00e2\u0080\u009d, the purchased company is Marshalls, not Company-B.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"4 70.4 46.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Several such particles may be prefixed onto a single stem, in which case the affixation is subject to strict linear precedence constraints.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Clearly, retaining the original frequencies is important for good performance, and globally smoothing the final weighted frequencies is crucial.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The weak hypothesis chosen was then restricted to be a predictor in favor of this label.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Recent work has made significant progress on unsupervised POS tagging (Me\u00c2\u00b4rialdo, 1994; Smith and Eisner, 2005; Haghighi and Klein, 2006; Johnson,2007; Goldwater and Griffiths, 2007; Gao and John son, 2008; Ravi and Knight, 2009).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Within the RST \u00e2\u0080\u009cuser community\u00e2\u0080\u009d there has also been discussion whether two levels of discourse structure should not be systematically distinguished (intentional versus informational).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We consider the unsupervised POS induction problem without the use of a tagging dictionary.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"A contextual rule considers words surrounding the string in the sentence in which it appears (e.g., a rule that any proper name modified by an appositive whose head is president is a person).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The human evaluators were a non-native, fluent Arabic speaker (the first author) for the ATB and a native English speaker for the WSJ.7 Table 5 shows type- and token-level error rates for each corpus.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Evaluation results within sets Table 1 shows the evaluation result based on the number of phrases in a set.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The reference medicine for Silapo is EPREX\/ERYPO, which contains epoetin alfa.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Section 2.2 then describes our representation for contextual roles and four types of contextual role knowledge that are learned from the training examples.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Two measures that can be used to compare judgments are: 1.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Under this scheme, n human judges are asked independently to segment a text.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Furthermore, by inverting the transducer so that it maps from phonemic transcriptions to hanzi sequences, one can apply the segmenter to other problems, such as speech recognition (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"37.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The hypothesized parse is then the set of constituents that are likely (P > 0.5) to be in the parse according to this model.","label":0},{"source_text":"Here we present two algorithms.","target_text":"For each label (Per s on, organization and Location), take the n contextual rules with the highest value of Count' (x) whose unsmoothed3 strength is above some threshold pmin.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Model Overview The model starts by generating a tag assignment T for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"We use a solution to this problem similar to the one presented in (Och et al., 1999), where target words are joined during training.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The paper is structured as follows.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Our empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We formulate the update as follows: where \u2200ui \u2208 Vf \\ Vfl, \u03b3i(y) and \u03bai are defined as: We ran this procedure for 10 iterations.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"collected too.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Fall.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The effect of UNIQ is very small on both data sets.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For a word whose initCaps might be due to its position rather than its meaning (in headlines, first word of a sentence, etc), the case information of other occurrences might be more accurate than its own.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"In the interest of testing the robustness of these combining techniques, we added a fourth, simple nonlexicalized PCFG parser.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Statistical methods seem particularly applicable to the problem of unknown-word identification, especially for constructions like names, where the linguistic constraints are minimal, and where one therefore wants to know not only that a particular se\u00c2\u00ad quence of hanzi might be a name, but that it is likely to be a name with some probabil\u00c2\u00ad ity.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As noted, this sentence consists of four words, namely B X ri4wen2 'Japanese,' :\u00c2\u00a5, zhanglyu2 'octopus\/ :&P:l zen3me0 'how,' and IDt shuol 'say.'","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: Mr. Bush disclosed the policy by reading it...","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"If evidence indicates that hypotheses C and D are less likely than hypotheses A and B, then probabilities are redistributed to reflect the fact that {A, B} is more likely to contain the answer than {C, D}.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The development of automatic scoring methods is an open field of research.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":", Sun day, then the feature DayOfTheWeek is set to 1.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"We respond to this on the one hand with a format for its underspecification (see 2.4) and on the other hand with an additional level of annotation that attends only to connectives and their scopes (see 2.5), which is intended as an intermediate step on the long road towards a systematic and objective treatment of rhetorical structure.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If it starts with a lower case letter, and contains both upper and lower case letters, then (mixedCaps, zone) is set to 1.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Intuitively, it places more weight on OUT when less evidence from IN is available.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Hereafter, each pair of NE categories will be called a domain; e.g. the \u00e2\u0080\u009cCompany \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain, which we will call CC- domain (Step 2).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Unsupervised Learning of Contextual Role Knowledge for Coreference Resolution","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"This approach makes the training objective more complex by adding linear constraints proportional to the number of word types, which is rather prohibitive.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"To eliminate such sequences, we define a transition probability between word classes to be equal to 1 if the sequence is admissible, and 0 otherwise.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We are interested in combining the substructures of the input parses to produce a better parse.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Tables 4 and 5 show BABAR\u00e2\u0080\u0099s performance when just one contextual role knowledge source is used at a time.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The development of the very first Hebrew Treebank (Sima\u2019an et al., 2001) called for the exploration of general statistical parsing methods, but the application was at first limited.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We experimented with increasingly rich grammars read off of the treebank.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In the third and final scheme, denoted Path, we keep the extra infor2Note that this is a baseline for the parsing experiment only (Experiment 2).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"If enough parsers suggest that a particular constituent belongs in the parse, we include it.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Our parsing performance measures (SY N) thus report the PARSEVAL extension proposed in Tsarfaty (2006).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For eight judges, ranging k between 1 and 8 corresponded to a precision score range of 90% to 30%, meaning that there were relatively few words (30% of those found by the automatic segmenter) on which all judges agreed, whereas most of the words found by the segmenter were such that one human judge agreed.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The segmenter will give both analyses :1 cai2 neng2 'just be able,' and ?]cai2neng2 'talent,' but the latter analysis is preferred since splitting these two morphemes is generally more costly than grouping them.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The techniques we develop can be extended in a relatively straightforward manner to the more general case when OUT consists of multiple sub-domains.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"However, using the top-level semantic classes of WordNet proved to be problematic because the class distinctions are too coarse.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Here, annotation proceeds in two phases: first, the domains and the units of IS are marked as such.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In Semitic languages the situation is very different.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The work of Thatcher (1973) and Rounds (1969) define formal systems that generate tree sets that are related to CFG's and IG's.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In these experiments, the input lacks segmentation markers, hence the slightly different dev set baseline than in Table 6.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Suppose M has to determine whether the k substrings ,.. .,ak can be derived from some symbol A.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Secondly, some segments in a proposed segment sequence may in fact be seen lexical events, i.e., for some p tag Prf(p \u2014* (s, p)) > 0, while other segments have never been observed as a lexical event before.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Our experimental setup therefore is designed to serve two goals.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"A receives a votes, and B receives b votes.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Person\u00e2\u0080\u009d domain, 618 keywords are found, and in the \u00e2\u0080\u009cCountry \u00e2\u0080\u0093 Country\u00e2\u0080\u009d domain, 303 keywords are found.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"rhetorical analysis We are experimenting with a hybrid statistical and knowledge-based system for discourse parsing and summarization (Stede 2003), (Hanneforth et al. 2003), again targeting the genre of commentaries.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Hence, s < k. But by addition of the votes on the two parses, s > 2N-11> k, a contradiction.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Keywords with more than one word In the evaluation, we explained that \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d are considered paraphrases.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"3 68.4 49.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The CFLex and CFNet knowledge sources provide positive evidence that a candidate NP and anaphor might be coreferent.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"One side of the decision making process is when we choose to believe a constituent should be in the parse, even though only one parser suggests it.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Also, we don\u00e2\u0080\u0099t know how many such paraphrase sets are necessary to cover even some everyday things or events.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"German English Training: Sentences 58 073 Words 519 523 549 921 Words* 418 979 453 632 Vocabulary Size 7939 4648 Singletons 3454 1699 Test-147: Sentences 147 Words 1 968 2 173 Perplexity { 26:5 Table 4: Multi-reference word error rate (mWER) and subjective sentence error rate (SSER) for three different search procedures.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"One is that smaller sets sometime have meaningless keywords, like \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d in the CC-domain, or \u00e2\u0080\u009ccompare\u00e2\u0080\u009d in the PC-domain.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The alignment mapping is j ! i = aj from source position j to target position i = aj . The use of this alignment model raises major problems if a source word has to be aligned to several target words, e.g. when translating German compound nouns.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The domains are the linguistic spans that are to receive an IS-partitioning, and the units are the (smaller) spans that can play a role as a constituent of such a partitioning.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Future work should also extend the approach to build a complete named entity extractor - a method that pulls proper names from text and then classifies them.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"First, the training data for the parser is projectivized by applying a minimal number of lifting operations (Kahane et al., 1998) and encoding information about these lifts in arc labels.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The terms on the right-hand-side denote the type-level and token-level probability terms respectively.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In total, for the 2,000 NE category pairs, 5,184 keywords are found.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Like verbs, maSdar takes arguments and assigns case to its objects, whereas it also demonstrates nominal characteristics by, e.g., taking determiners and heading iDafa (Fassi Fehri, 1993).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Unfortunately, the best completely unsupervised English POS tagger (that does not make use of a tagging dictionary) reaches only 76.1% accuracy (Christodoulopoulos et al., 2010), making its practical usability questionable at best.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"This decreases the statistical significance of our results compared to those studies.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Then, token- level HMM emission parameters are drawn conditioned on these assignments such that each word is only allowed probability mass on a single assigned tag.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Time includes all queries but excludes random number generation and data structure population.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"On the English side, however, the vertices (denoted by Ve) correspond to word types.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"This suggests that by generalizing the notion of context-freeness in CFG's, we can define a class of grammatical formalisms that manipulate more complex structures.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"This problem arises because our keywords consist of only one word.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The key point is that the second constraint can be remarkably powerful in reducing the complexity of the learning problem.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Cohen and Smith (2007) followed up on these results and proposed a system for joint inference of morphological and syntactic structures using factored models each designed and trained on its own.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"To explicitly handle the word reordering between words in source and target language, we use the concept of the so-called inverted alignments as given in (Ney et al., 2000).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Sheffield Guthrie and Hepple (2010) explore several randomized compression techniques, but did not release code.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As noted, this sentence consists of four words, namely B X ri4wen2 'Japanese,' :\u00c2\u00a5, zhanglyu2 'octopus\/ :&P:l zen3me0 'how,' and IDt shuol 'say.'","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"pronunciation depends upon word affiliation: tfJ is pronounced deO when it is a prenominal modification marker, but di4 in the word \u00c2\u00a7tfJ mu4di4 'goal'; fl; is normally ganl 'dry,' but qian2 in a person's given name.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The method halves the error rate in comparison to a method using the labeled examples alone.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The weak learner for two-class problems computes a weak hypothesis h from the input space into the reals (h : 2x -4 R), where the sign4 of h(x) is interpreted as the predicted label and the magnitude I h(x)I is the confidence in the prediction: large numbers for I h(x)I indicate high confidence in the prediction, and numbers close to zero indicate low confidence.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For each pair of judges consider one judge as the standard,.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We are not claiming that this method is almighty.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Since our destructor is an efficient call to munmap, bypassing the destructor favors only other packages.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Formally, let el (62) be the number of classification errors of the first (second) learner on the training data, and let Eco be the number of unlabeled examples on which the two classifiers disagree.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The larger sets are more accurate than the small sets.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"First, the model assumes independence between the first and second hanzi of a double given name.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Either save money at any cost - or give priority to education.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Note that these observa sider suffix features, capitalization features, punctuation, and digit features.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"N, portion of examples on which both classifiers give a label rather than abstaining), and the proportion of these examples on which the two classifiers agree.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"When a collision occurs, linear probing places the entry to be inserted in the next (higher index) empty bucket, wrapping around as necessary.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"\u201cin the shadow\u201d.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"For example, we can easily imagine that the number of paraphrases for \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d is enormous and it is not possible to create a comprehensive inventory by hand.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Unlike Germann et al. (2009), we chose a model size so that all benchmarks fit comfortably in main memory.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Some relations are signalled by subordinating conjunctions, which clearly demarcate the range of the text spans related (matrix clause, embedded clause).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Note that the good classes JADE, GOLD and GRASS have lower costs than the bad classes SICKNESS, DEATH and RAT, as desired, so the trend observed for the results of this method is in the right direction.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The 95% confidence intervals for type-level errors are (5580, 9440) for the ATB and (1400, 4610) for the WSJ.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The weak hypothesis can abstain from predicting the label of an instance x by setting h(x) = 0.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"This drastic tree manipulation is not appropriate for situations in which we want to assign particular structures to sentences.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Other strategies could readily 6 As a reviewer has pointed out, it should be made clear that the function for computing the best path is. an instance of the Viterbi algorithm.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This leads to word- and constituent-boundaries discrepancy, which breaks the assumptions underlying current state-of-the-art statistical parsers.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"In practice, this sparsity constraint is difficult to incorporate in a traditional POS induction system (Me\u00c2\u00b4rialdo, 1994; Johnson, 2007; Gao and Johnson, 2008; Grac\u00c2\u00b8a et al., 2009; Berg-Kirkpatrick et al., 2010).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"21 In Chinese, numerals and demonstratives cannot modify nouns directly, and must be accompanied by.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"As an indication, in our core corpus, we found an average sentence length of 15.8 words and 1.8 verbs per sentence, whereas a randomly taken sample of ten commentaries from the national papers Su\u00c2\u00a8ddeutsche Zeitung and Frankfurter Allgemeine has 19.6 words and 2.1 verbs per sentence.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Ex: Mr. Bush disclosed the policy by reading it...","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Second, rather than relying on a division of the corpus into manually-assigned portions, we use features intended to capture the usefulness of each phrase pair.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Chinese word segmentation can be viewed as a stochastic transduction problem.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"AdaBoost.MH maintains a distribution over instances and labels; in addition, each weak-hypothesis outputs a confidence vector with one confidence value for each possible label.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"5 70.1 58.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Consequently, we cut their evidence values in half to lessen their influence.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The task can be considered to be one component of the MUC (MUC-6, 1995) named entity task (the other task is that of segmentation, i.e., pulling possible people, places and locations from text before sending them to the classifier).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Even if an example like this is not labeled, it can be interpreted as a &quot;hint&quot; that Mr and president imply the same category.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"diesem 3.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"For developing these mechanisms, the possibility to feed in hand-annotated information is very useful.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The following features were used: full-string=x The full string (e.g., for Maury Cooper, full- s tring=Maury_Cooper). contains(x) If the spelling contains more than one word, this feature applies for any words that the string contains (e.g., Maury Cooper contributes two such features, contains (Maury) and contains (Cooper) . allcapl This feature appears if the spelling is a single word which is all capitals (e.g., IBM would contribute this feature). allcap2 This feature appears if the spelling is a single word which is all capitals or full periods, and contains at least one period.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In this paper k = 3 (the three labels are person, organization, location), and we set a = 0.1.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Discriminative Instance Weighting for Domain Adaptation in Statistical Machine Translation","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"This allow the learners to &quot;bootstrap&quot; each other by filling the labels of the instances on which the other side has abstained so far.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Making the reasonable assumption that similar information is relevant for solving these problems in Chinese, it follows that a prerequisite for intonation-boundary assignment and prominence assignment is word segmentation.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Unsupervised Models for Named Entity Classification Collins","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Table 8a shows that the best model recovers SBAR at only 71.0% F1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Evaluation of Morphological Analysis.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Consider the case where IX].","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Roughly speaking, a language, L, has the property of semilinearity if the number of occurrences of each symbol in any string is a linear combination of the occurrences of these symbols in some fixed finite set of strings.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The corresponding token words w are drawn conditioned on t and \u00ce\u00b8.2 Our full generative model is given by: K P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) = n (P (\u00cf\u0086t|\u00ce\u00b1)P (\u00ce\u00b8t|T , \u00ce\u00b1)) t=1 The transition distribution \u00cf\u0086t for each tag t is drawn according to DIRICHLET(\u00ce\u00b1, K ), where \u00ce\u00b1 is the shared transition and emission distribution hyperparameter.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Acknowledgments We thank Meni Adler and Michael Elhadad (BGU) for helpful comments and discussion.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Then, moving from connective to connective, ConAno sometimes offers suggestions for its scope (using heuristics like \u00e2\u0080\u0098for sub- junctor, mark all words up to the next comma as the first segment\u00e2\u0080\u0099), which the annotator can accept with a mouseclick or overwrite, marking instead the correct scope with the mouse.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"When the connective is an adverbial, there is much less clarity as to the range of the spans.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The task is to learn a function from an input string (proper name) to its type, which we will assume to be one of the categories Person, Organization, or Location.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"A graphical depiction of our model as well as a summary of random variables and parameters can be found in Figure 1.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each resolution in the training data, BABAR also associates the co-referring expression of an NP with the NP\u00e2\u0080\u0099s caseframe.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Sie.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In MUC6 and MUC7, the named entity task is defined as finding the following classes of names: person, organization, location, date, time, money, and percent (Chinchor, 1998; Sundheim, 1995) Machine learning systems in MUC6 and MUC 7 achieved accuracy comparable to rule-based systems on the named entity task.","label":0},{"source_text":"There is no global pruning.","target_text":"The Levenshtein distance between the automatic translation and each of the reference translations is computed, and the minimum Levenshtein distance is taken.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"There is a sizable literature on Chinese word segmentation: recent reviews include Wang, Su, and Mo (1990) and Wu and Tseng (1993).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Full Chinese personal names are in one respect simple: they are always of the form family+given.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The same numbers were used for each data structure.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Such tag sequences are often treated as \u201ccomplex tags\u201d (e.g.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Automatic paraphrase discovery is an important but challenging task.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"\u00e2\u0080\u0094ested in correlations between prosody and dis course structure.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In this section, we describe how contextual role knowledge is represented and learned.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"For rules p : A fpo such that fp is constant function, giving an elementary structure, fp is defined such that fp() = (Si ... xi() where each z is a constant string.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Output of the learning algorithm: a function h:Xxy [0, 1] where h(x, y) is an estimate of the conditional probability p(y1x) of seeing label y given that feature x is present.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"(we ignored the 419 trees in their development set.)","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This is a rather important source of errors in name identifi\u00c2\u00ad cation, and it is not really possible to objectively evaluate a name recognition system without considering the main lexicon with which it is used.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In Modern Hebrew (Hebrew), a Semitic language with very rich morphology, particles marking conjunctions, prepositions, complementizers and relativizers are bound elements prefixed to the word (Glinert, 1989).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Although matching is done at the sentence level, this information is subsequently discarded when all matches are pooled.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We use a simple TF\/IDF method to measure the topicality of words.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"For example, we may have semantic information (e.g. database query operations) associated with the productions in a grammar.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Juri Ganitkevitch answered questions about Joshua.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"All three curves remain steep at the maximum training set size of 18818 trees.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Memory usage is likely much lower than ours. fThe original paper (Germann et al., 2009) provided only 2s of query timing and compared with SRI when it exceeded available RAM.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If so, the CF Network reports that the anaphor and candidate may be coreferent.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We define the lattice L to be the concatenation of the lattices Li corresponding to the input words wi (s.t.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","label":1},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This line of work has been motivated by empirical findings that the standard EM-learned unsupervised HMM does not exhibit sufficient word tag sparsity.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Daum\u00b4e (2007) applies a related idea in a simpler way, by splitting features into general and domain-specific versions.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The intuition here is that the role of a discourse marker can usually be de 9 Both the corpus split and pre-processing code are avail-.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Although these are technically nominal, they have become known as \u00e2\u0080\u009cequational\u00e2\u0080\u009d sentences.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Given that weights on all outgoing arcs sum up to one, weights induce a probability distribution on the lattice paths.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"input token, the segmentation is then performed deterministically given the 1-best analysis.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We evaluate our approach on eight European languages (\u00a76), and show that both our contributions provide consistent and statistically significant improvements.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"However, the accuracy is considerably higher than previously reported results for robust non-projective parsing of Czech, with a best performance of 73% UAS (Holan, 2004).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Hence, for each token, one of the four features zone-TXT, zone- HL, zone-DATELINE, or zone-DD is set to 1, and the other 3 are set to 0.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Using RandLM and the documented settings (8-bit values and 1 256 false-positive probability), we built a stupid backoff model on the same data as in Section 5.2.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"In line with perplexity results from Table 1, the PROBING model is the fastest followed by TRIE, and subsequently other packages.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The state function is integrated into the query process so that, in lieu of the query p(wnjwn\u22121 1 ), the application issues query p(wnjs(wn\u22121 1 )) which also returns s(wn1 ).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Conditioned on T , features of word types W are drawn.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Relative pronouns with only 1 NP in scope..","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Because of their size, the examples (Figures 2 to 4) appear at the end of the paper.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"For developers of Statistical Machine Translation (SMT) systems, an additional complication is the heterogeneous nature of SMT components (word-alignment model, language model, translation model, etc.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The results, along with the total number of phrases, are shown in Table 1.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Whereas Arabic linguistic theory as Saddam (a) Reference (b) Stanford signs (1) and (2) to the class of pseudo verbs 01 +i J>1\u00ef\u00bf\u00bd inna and her sisters since they can beinflected, the ATB conventions treat (2) as a com plementizer, which means that it must be the head of SBAR.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"3.1 Lexicon Component.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"In our particular case the majority requires the agreement of only two parsers because we have only three.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The translation scores for the hypotheses generated with different threshold values t0 are compared to the translation scores obtained with a conservatively large threshold t0 = 10:0 . For each test series, we count the number of sentences whose score is worse than the corresponding score of the test series with the conservatively large threshold t0 = 10:0, and this number is reported as the number of search errors.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Instead, the designs of the various annotation layers and the actual annotation work are results of a series of diploma theses, of students\u00e2\u0080\u0099 work in course projects, and to some extent of paid assistentships.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Figure 2 shows examples of extracted NE pair instances and their contexts.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"An ATM has two types of states, existential and universal.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Feature-based HMM Model (Berg- Kirkpatrick et al., 2010): The KM model uses a variety of orthographic features and employs the EM or LBFGS optimization algorithm; Posterior regulariation model (Grac\u00c2\u00b8a et al., 2009): The G10 model uses the posterior regular- ization approach to ensure tag sparsity constraint.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For Arabic we M o d e l S y s t e m L e n g t h L e a f A n c e s t o r Co rpu s Sent Exact E v a l b L P LR F1 T a g % B a s e l i n e 7 0 St an for d (v 1.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We model po(s|t) using a MAP criterion over weighted phrase-pair counts: and from the similarity to (5), assuming y = 0, we see that w\u03bb(s, t) can be interpreted as approximating pf(s, t)\/po(s, t).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Roughly speaking, the new algorithm presented in this paper performs a similar search, but instead minimizes a bound on the number of (unlabeled) examples on which two classifiers disagree.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Twentieth-century linguistic work on Chinese (Chao 1968; Li and Thompson 1981; Tang 1988,1989, inter alia) has revealed the incorrectness of this traditional view.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Dagan and Itai (Dagan and Itai, 1990) experimented with co-occurrence statistics that are similar to our lexical caseframe expectations.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Thatcher (1973) describes a tee pumping lemma for recognizable sets related to the string pumping lemma for regular sets.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"17 They also provide a set of title-driven rules to identify names when they occur before titles such as $t. 1: xianlshengl 'Mr.' or i:l:itr!J tai2bei3 shi4zhang3 'Taipei Mayor.'","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"First, the training data for the parser is projectivized by applying a minimal number of lifting operations (Kahane et al., 1998) and encoding information about these lifts in arc labels.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Furthermore we expect the label distributions on the foreign to be fairly noisy, because the graph constraints have not been taken into account yet.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"To formalize the approach, we introduce four verbgroup states S: Initial (I): A contiguous, initial block of source positions is covered.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Searching a probing hash table consists of hashing the key, indexing the corresponding bucket, and scanning buckets until a matching key is found or an empty bucket is encountered, in which case the key does not exist in the table.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Alon Lavie advised on this work.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"3.1 Gross Statistics.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"16 As one reviewer points out, one problem with the unigram model chosen here is that there is still a. tendency to pick a segmentation containing fewer words.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"An input ABCD can be represented as an FSA as shown in Figure 2(b).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Table 8b shows that verbal nouns are the hardest pre-terminal categories to identify.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If no candidate satisfies this condition (which is often the case), then the anaphor is left unresolved.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Cur In order to ground such approaches in linguistic observation and description, a multi-level anno 10 For an exposition of the idea as applied to the task of text planning, see (Chiarcos, Stede 2004).","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The first experiment uses data from two dependency treebanks.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"(S2) The burglar fired the gun three times and fled.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We also removed the context-type feature type when using the CoBoost approach.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Similarly, the tokens and are tested against each list, and if found, a corresponding feature will be set to 1.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"This approach makes the training objective more complex by adding linear constraints proportional to the number of word types, which is rather prohibitive.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We found that contextual role knowledge was more beneficial for pronouns than for definite noun phrases.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"On the other hand, only very restricted reorderings are necessary, e.g. for the translation direction from Table 2: Coverage set hypothesis extensions for the IBM reordering.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Unlike Kahane et al. (1998), we do not regard a projectivized representation as the final target of the parsing process.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"If the semantic expectations do not intersect, then we know that the case- frames extract mutually exclusive types of noun phrases.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The rationale for treating these semantic labels differently is that they are specific and reliable (as opposed to the WordNet classes, which are more coarse and more noisy due to polysemy).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Previous reports on Chinese segmentation have invariably cited performance either in terms of a single percent-correct score, or else a single precision-recall pair.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Also, the method of using keywords rules out phrases which don\u00e2\u0080\u0099t contain popular words in the domain.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We obtained positive results using a very simple phrase-based system in two different adaptation settings: using English\/French Europarl to improve a performance on a small, specialized medical domain; and using non-news portions of the NIST09 training material to improve performance on the news-related corpora.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"This line of work has been motivated by empirical findings that the standard EM-learned unsupervised HMM does not exhibit sufficient word tag sparsity.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The state function is integrated into the query process so that, in lieu of the query p(wnjwn\u22121 1 ), the application issues query p(wnjs(wn\u22121 1 )) which also returns s(wn1 ).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Therefore, for n-gram wn1 , all leftward extensions wn0 are an adjacent block in the n + 1-gram array.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Both of the switching techniques, as well as the parametric hybridization technique were also shown to be robust when a poor parser was introduced into the experiments.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The joint morphological and syntactic hypothesis was first discussed in (Tsarfaty, 2006; Tsarfaty and Sima\u2019an, 2004) and empirically explored in (Tsarfaty, 2006).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Any NLP application that presumes as input unrestricted text requires an initial phase of text analysis; such applications involve problems as diverse as machine translation, information retrieval, and text-to-speech synthesis (TIS).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Dictionaries: Due to the limited amount of training material, name dictionaries have been found to be useful in the named entity task.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"76 16.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"With some minor API changes, namely returning the length of the n-gram matched, it could also be faster\u2014though this would be at the expense of an optimization we explain in Section 4.1.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The composition operations in the case of CFG's are parameterized by the productions.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In this section, we will explain the algorithm step by step with examples.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The same form fmnh can be segmented as f-mnh, f (\u201cthat\u201d) functioning as a reletivizer with the form mnh.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"BABAR uses information extraction patterns to identify contextual roles and creates four contextual role knowledge sources using unsupervised learning.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In these examples, the names identified by the two systems (if any) are underlined; the sentence with the correct segmentation is boxed.19 The differences in performance between the two systems relate directly to three issues, which can be seen as differences in the tuning of the models, rather than repre\u00c2\u00ad senting differences in the capabilities of the model per se.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"html 4 www.wagsoft.com\/RSTTool assigning rhetorical relations is a process loaded with ambiguity and, possibly, subjectivity.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"When 7' is adjoined at ?I in the tree 7 we obtain a tree v&quot;.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"where the husband's family name is optionally prepended to the woman's full name; thus ;f:*lf#i xu3lin2-yan2hai3 would represent the name that Ms. Lin Yanhai would take if she married someone named Xu.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Thus in an English sentence such as I'm going to show up at the ACL one would reasonably conjecture that there are eight words separated by seven spaces.","label":0},{"source_text":"There is no global pruning.","target_text":"We can do that . Input: Das ist zu knapp , weil ich ab dem dritten in Kaiserslautern bin . Genaugenommen nur am dritten . Wie ware es denn am ahm Samstag , dem zehnten Februar ? MonS: That is too tight , because I from the third in Kaiserslautern . In fact only on the third . How about ahm Saturday , the tenth of February ? QmS: That is too tight , because I am from the third in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? IbmS: That is too tight , from the third because I will be in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? Input: Wenn Sie dann noch den siebzehnten konnten , ware das toll , ja . MonS: If you then also the seventeenth could , would be the great , yes . QmS: If you could then also the seventeenth , that would be great , yes . IbmS: Then if you could even take seventeenth , that would be great , yes . Input: Ja , das kommt mir sehr gelegen . Machen wir es dann am besten so . MonS: Yes , that suits me perfectly . Do we should best like that . QmS: Yes , that suits me fine . We do it like that then best . IbmS: Yes , that suits me fine . We should best do it like that .","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"To provide a thorough analysis, we evaluated three baselines and two oracles in addition to two variants of our graph-based approach.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"For each language and setting, we report one-to-one (11) and many- to-one (m-1) accuracies.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"(a) Of the high frequency phrasal categories, ADJP and SBAR are the hardest to parse.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Asking the annotator to also formulate the question is a way of arriving at more reproducible decisions.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"annotation guidelines that tell annotators what to do in case of doubt.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"For example, take ..., says Maury Cooper, a vice president at S.&P.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Chris Dyer integrated the code into cdec.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The idea is to have a pipeline of shallow-analysis modules (tagging, chunk- ing, discourse parsing based on connectives) and map the resulting underspecified rhetorical tree (see Section 2.4) into a knowledge base that may contain domain and world knowledge for enriching the representation, e.g., to resolve references that cannot be handled by shallow methods, or to hypothesize coherence relations.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"As noted in Section 1, our code finds the longest matching entry wnf for query p(wn|s(wn\u22121 f ) The probability p(wn|wn\u22121 f ) is stored with wnf and the backoffs are immediately accessible in the provided state s(wn\u22121 When our code walks the data structure to find wnf , it visits wnn, wnn\u22121, ... , wnf .","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"At present, the \u00e2\u0080\u0098Potsdam Commentary Corpus\u00e2\u0080\u0099 (henceforth \u00e2\u0080\u0098PCC\u00e2\u0080\u0099 for short) consists of 170 commentaries from Ma\u00c2\u00a8rkische Allgemeine Zeitung, a German regional daily.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The 3rd block contains the mixture baselines.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"A novel element of our model is the ability to capture type-level tag frequencies.","label":0},{"source_text":"There is no global pruning.","target_text":"This algorithm can be applied to statistical machine translation.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This offers the well-known advantages for inter- changability, but it raises the question of how to query the corpus across levels of annotation.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Each connected path (l1 ... lk) E L corresponds to one morphological segmentation possibility of W. The Parser Given a sequence of input tokens W = w1 ... wn and a morphological analyzer, we look for the most probable parse tree \u03c0 s.t.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"(Bar-Haim et al., 2007; Habash and Rambow, 2005)) and probabilities are assigned to different analyses in accordance with the likelihood of their tags (e.g., \u201cfmnh is 30% likely to be tagged NN and 70% likely to be tagged REL+VB\u201d).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"5 60.6 Table 3: Multilingual Results: We report token-level one-to-one and many-to-one accuracy on a variety of languages under several experimental settings (Section 5).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We conjecture that this trend may continue by incorporating additional information, e.g., three-dimensional models as proposed by Tsarfaty and Sima\u2019an (2007).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"To measure the contribution of each modification, a third, intermediate algorithm, Yarowsky-cautious was also tested.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Since these path sets are semilinear, the property of independent paths in their tree sets is sufficient to cause semilinearity of the languages generated by them.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In the first part of the experiment, dependency graphs from the treebanks were projectivized using the algorithm described in section 2.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thus, M works in logspace and recognition can be done on a deterministic TM in polynomial tape.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"We use a solution to this problem similar to the one presented in (Och et al., 1999), where target words are joined during training.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"of Articles No.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Our new approach, which is called quasi-monotone search, processes the source sentence monotonically, while explicitly taking into account the positions of the German verbgroup.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"However, if we consider precision, recall and Fmeasure on non-projective dependencies only, as shown in Table 6, some differences begin to emerge.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"During training, we treat as observed the language word types W as well as the token-level corpus w. We utilize Gibbs sampling to approximate our collapsed model posterior: P (T ,t|W , w, \u00ce\u00b1, \u00ce\u00b2) \u00e2\u0088\u009d P (T , t, W , w|\u00ce\u00b1, \u00ce\u00b2) 0.7 0.6 0.5 0.4 0.3 English Danish Dutch Germany Portuguese Spanish Swedish = P (T , t, W , w, \u00cf\u0088, \u00ce\u00b8, \u00cf\u0086, w|\u00ce\u00b1, \u00ce\u00b2)d\u00cf\u0088d\u00ce\u00b8d\u00cf\u0086 Note that given tag assignments T , there is only one setting of token-level tags t which has mass in the above posterior.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We hypothesize that modeling morphological information will greatly constrain the set of possible tags, thereby further refining the representation of the tag lexicon.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We simulate lexical constraints by using an external lexical resource against which we verify whether OOV segments are in fact valid Hebrew lexemes.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The weight on each sentence is a value in [0, 1] computed by a perceptron with Boolean features that indicate collection and genre membership.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Pseudo-code describing the generalized boosting algorithm of Schapire and Singer is given in Figure 1.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Microsoft\u2019s approach uses dependency trees, others use hierarchical phrase models.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The state in future has not enough work for its many teachers.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Training under this model involves estimation of parameter values for P(y), P(m) and P(x I y).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Supervised learning approaches have advanced the state-of-the-art on a variety of tasks in natural language processing, resulting in highly accurate systems.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":".t The path set of tree sets at level k +1 have the complexity of the string language of level k. The independence of paths in a tree set appears to be an important property.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Clearly the percentage of productively formed words is quite small (for this particular corpus), meaning that dictionary entries are covering most of the 15 GR is .73 or 96%..","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Section 3 discusses the applications that have been completed with PCC, or are under way, or are planned for the future.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In total 13,976 phrases are assigned to sets of phrases, and the accuracy on our evaluation data ranges from 65 to 99%, depending on the domain and the size of the sets.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"These two principles guide experimentation in this framework, and together with the evaluation measures help us decide which specific type of substructure to combine.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We call this approach parse hybridization.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"At each point during the derivation, the prediction is based on six word tokens, the two topmost tokens on the stack, and the next four input tokens.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The interdependence between fb or 1\/!i, and 5:2 is not captured by our model, but this could easily be remedied.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The original tag set for the CoNLL-X Dutch data set consists of compounded tags that are used to tag multi-word units (MWUs) resulting in a tag set of over 300 tags.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Finally, we would like to note that it is possible to devise similar algorithms based with other objective functions than the one given in Equ.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In addition, there are several approaches to non-projective dependency parsing that are still to be evaluated in the large (Covington, 1990; Kahane et al., 1998; Duchier and Debusmann, 2001; Holan et al., 2001; Hellwig, 2003).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"3.2 Stochastic rhetorical analysis.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We construct a mapping from all the space-delimited tokens seen in the training sentences to their corresponding analyses.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We show that noun-noun vs. discourse-level coordination ambiguity in Arabic is a significant source of parsing errors (Table 8c).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"For example, in .., says Mr. Cooper, a vice president of.. both a spelling feature (that the string contains Mr.) and a contextual feature (that president modifies the string) are strong indications that Mr. Cooper is of type Person.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"It was developed in response to the non-terminal\/terminal bias of Evalb, but Clegg and Shepherd (2005) showed that it is also a valuable diagnostic tool for trees with complex deep structures such as those found in the ATB.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"These methods demonstrated the benefits of incorporating linguistic features using a log-linear parameterization, but requires elaborate machinery for training.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The gender, number, and scoping KSs eliminate candidates from consideration.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The final model tions.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Note that Zt is a normalization constant that ensures the distribution Dt+i sums to 1; it is a function of the weak hypothesis ht and the weight for that hypothesis at chosen at the tth round.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In IdentiFinder, there is a priority in the feature assignment, such that if one feature is used for a token, another feature lower in priority will not be used.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We first note that the accuracy results of our system are overall higher on their setup, on all measures, indicating that theirs may be an easier dataset.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(7) is at 0 when: 1) Vi : sign(gi (xi)) = sign(g2 (xi)); 2) Ig3(xi)l oo; and 3) sign(gi (xi)) = yi for i = 1, , m. In fact, Zco provides a bound on the sum of the classification error of the labeled examples and the number of disagreements between the two classifiers on the unlabeled examples.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The number of latent HMM states for each language in our experiments was set to the number of fine tags in the language\u2019s treebank.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In the following, we use the notation wi wj to mean that (wi, r, wj) E A; r we also use wi wj to denote an arc with unspecified label and wi \u2014*\u2217 wj for the reflexive and transitive closure of the (unlabeled) arc relation.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"tai2du2 'Taiwan Independence.'","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"(f1; ;mg n fl1g ; l) 3 (f1; ;mg n fl; l1; l2g ; l0) !","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Similarly, hanzi sharing the GHOST radical _m tend to denote spirits and demons, such as _m gui3 'ghost' itself, II: mo2 'demon,' and yan3 'nightmare.'","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"85 82.","label":0},{"source_text":"Here we present two algorithms.","target_text":"A spelling rule might be a simple look-up for the string (e.g., a rule that Honduras is a location) or a rule that looks at words within a string (e.g., a rule that any string containing Mr. is a person).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Our results show that MENERGI performs as well as IdentiFinder when trained on comparable amount of training data.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"(2009).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Examples are given in Table 4.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Across all languages, +PRIOR consistently outperforms 1TW, reducing error on average by 9.1% and 5.9% on best and median settings respectively.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"E.g. when 'Zahnarzttermin' is aligned to dentist's, the extended lexicon model might learn that 'Zahnarzttermin' actuallyhas to be aligned to both dentist's and ap pointment.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The phrases have to be the expressions of length less than 5 chunks, appear between two NEs.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We tokenize MWUs and their POS tags; this reduces the tag set size to 12.","label":0},{"source_text":"There is no global pruning.","target_text":"For \u00c3\u0086 = 0, no new target word is generated, while an additional source sentence position is covered.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Fall.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Thus, the language generated by a grammar of a LCFRS is semilinear.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"As suggested in Section 4.3.2, a derivation with independent paths can be divided into subcomputations with limited sharing of information.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"72 78.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In a few cases, the criteria for correctness are made more explicit.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"But gold segmentation is not available in application settings, so a segmenter and parser are arranged in a pipeline.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"These results are achieved by training on the official MUC6 and MUC7 training data, which is much less training data than is used by other machine learning systems that worked on the MUC6 or MUC7 named entity task (Bikel et al., 1997; Bikel et al., 1999; Borth- wick, 1999).","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"For each token, three types of features may be taken into account: the word form; the part-of-speech assigned by an automatic tagger; and labels on previously assigned dependency arcs involving the token \u2013 the arc from its head and the arcs to its leftmost and rightmost dependent, respectively.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We adopted this state-of-the-art model because it makes it easy to experiment with various ways of incorporating our novel constraint feature into the log-linear emission model.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"BABAR uses two methods to identify anaphors that can be easily and reliably resolved with their antecedent: lexical seeding and syntactic seeding.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Nonetheless, the results of the comparison with human judges demonstrates that there is mileage being gained by incorporating models of these types of words.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The system described in this paper is similar to the MENE system of (Borthwick, 1999).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"= p(fj je) max \u00c3\u0086;e00 j02Cnfjg np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j 0 )o: The DP equation is evaluated recursively for each hypothesis (e0; e; C; j).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"When SRILM estimates a model, it sometimes removes n-grams but not n + 1-grams that extend it to the left.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Since each zi is a contiguous substring of the input (say ai,), and no two substrings overlap, we can represent zi by the pair of integers (i2, i2).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"GL is then used to parse the string tn1 ... tnk_1, where tni is a terminal corresponding to the lattice span between node ni and ni+1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Clearly this is not the only way to estimate word-frequencies, however, and one could consider applying other methods: in partic\u00c2\u00ad ular since the problem is similar to the problem of assigning part-of-speech tags to an untagged corpus given a lexicon and some initial estimate of the a priori probabilities for the tags, one might consider a more sophisticated approach such as that described in Kupiec (1992); one could also use methods that depend on a small hand-tagged seed corpus, as suggested by one reviewer.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We used the TF\/ITF metric to identify keywords.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Roughly speaking, previous work can be divided into three categories, namely purely statistical approaches, purely lexi\u00c2\u00ad cal rule-based approaches, and approaches that combine lexical information with sta\u00c2\u00ad tistical information.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This has the effect of randomly permuting vocabulary identifiers, meeting the requirements of interpolation search when vocabulary identifiers are used as keys.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This combination generalizes (2) and (3): we use either at = a to obtain a fixed-weight linear combination, or at = cI(t)\/(cI(t) + 0) to obtain a MAP combination.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Also, the argument has been made that machine translation performance should be evaluated via task-based evaluation metrics, i.e. how much it assists performing a useful task, such as supporting human translators or aiding the analysis of texts.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Only one of the first n positions which are not already aligned in a partial hypothesis may be chosen, where n is set to 4.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Next, we represent the input sentence as an unweighted finite-state acceptor (FSA) I over H. Let us assume the existence of a function Id, which takes as input an FSA A, and produces as output a transducer that maps all and only the strings of symbols accepted by A to themselves (Kaplan and Kay 1994).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We are interested in combining the substructures of the input parses to produce a better parse.","label":0},{"source_text":"There is no global pruning.","target_text":"The word joining is done on the basis of a likelihood criterion.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The rhetorical structure annotations of PCC have all been converted to URML.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The first four affixes are so-called resultative affixes: they denote some prop\u00c2\u00ad erty of the resultant state of a verb, as in E7 wang4bu4-liao3 (forget-not-attain) 'cannot forget.'","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"It uses a maximum entropy framework and classifies each word given its features.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(5) and ht into Equ.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"For the English RST-annotated corpus that is made available via LDC, his corresponding result is 62%.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Smith estimates Lotus will make profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"(b) After they were released...","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Specifically, the lexicon is generated as: P (T , W |\u00cf\u0088) =P (T )P (W |T ) Word Type Features (FEATS): Past unsupervised POS work have derived benefits from features on word types, such as suffix and capitalization features (Hasan and Ng, 2009; Berg-Kirkpatrick et al.,2010).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"As can be seen in Table 4, our training data is a lot less than those used by MENE and IdentiFinder3.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The features we used can be divided into 2 classes: local and global.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"If two phrases can be used to express the same relationship within an information extraction application (\u00e2\u0080\u009cscenario\u00e2\u0080\u009d), these two phrases are paraphrases.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The knowledge base then can be tested for its relation-inference capabilities on the basis of full-blown co-reference information.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"6 Results and Analysis.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The normalization factor plays an important role in the AdaBoost algorithm.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"In application settings, this may be a profitable strategy.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Based on these high-confidence alignments we can extract tuples of the form [u H v], where u is a foreign trigram type, whose middle word aligns to an English word type v. Our bilingual similarity function then sets the edge weights in proportion to these tuple counts.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The baseline system in Table 3 refers to the maximum entropy system that uses only local features.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Equ.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"More examples are shown in Figure 5.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"To differentiate between the coordinating and discourse separator functions of conjunctions (Table 3), we mark each CC with the label of its right sister (splitCC).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"43 80.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"If enough parsers suggest that a particular constituent belongs in the parse, we include it.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"7 Conclusion and Future Work.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Also there are cases where one of the two NEs belong to a phrase outside of the relation.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Two lists, Corporate-Suffix-List (for corporate suffixes) and Person-Prefix-List (for person prefixes), are collected from the training data.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The ATB has a much higher fraction of nuclei per tree, and a higher type-level error rate.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Models To assess the marginal utility of each component of the model (see Section 3), we incremen- tally increase its sophistication.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Output of the learning algorithm: a function h:Xxy [0, 1] where h(x, y) is an estimate of the conditional probability p(y1x) of seeing label y given that feature x is present.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Queries take the form p(wn|wn\u22121 1 ) where wn1 is an n-gram.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Then, the German infinitive 'besuchen' and the negation particle 'nicht' are translated.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The model we use provides a simple framework in which to incorporate a wide variety of lexical information in a uniform way.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"MENE without Proteus, however, did not do very well and only achieved an F measure of 84.22% (Borthwick, 1999).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"4.5 Transliterations of Foreign Words.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The overall performance of the LTG system was outstanding, but the system consists of a sequence of many hand-coded rules and machine-learning modules.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Because the information flow in our graph is asymmetric (from English to the foreign language), we use different types of vertices for each language.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The first probability is estimated from a name count in a text database, and the rest of the probabilities are estimated from a large list of personal names.n Note that in Chang et al.'s model the p(rule 9) is estimated as the product of the probability of finding G 1 in the first position of a two-hanzi given name and the probability of finding G2 in the second position of a two-hanzi given name, and we use essentially the same estimate here, with some modifications as described later on.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Given names are most commonly two hanzi long, occasionally one hanzi long: there are thus four possible name types, which can be described by a simple set of context-free rewrite rules such as the following: 1.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The POS distributions over the foreign trigram types are used as features to learn a better unsupervised POS tagger (\u00a75).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"On one end of the spectrum are clustering approaches that assign a single POS tag to each word type (Schutze, 1995; Lamar et al., 2010).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, the company may refer to Company X in one paragraph and Company Y in another.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We experiment with four values for each hyperparameter resulting in 16 (\u00ce\u00b1, \u00ce\u00b2) combinations: \u00ce\u00b1 \u00ce\u00b2 0.001, 0.01, 0.1, 1.0 0.01, 0.1, 1.0, 10 Iterations In each run, we performed 30 iterations of Gibbs sampling for the type assignment variables W .4 We use the final sample for evaluation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The relevance of the distinction between, say, phonological words and, say, dictionary words is shown by an example like rpftl_A :;!:Hfllil zhong1hua2 ren2min2 gong4he2-guo2 (China people republic) 'People's Republic of China.'","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We present several variations for the lexical component P (T , W |\u00cf\u0088), each adding more complex pa rameterizations.","label":0},{"source_text":"There is no global pruning.","target_text":"Search t0 CPU time #search mWER Method [sec] error [%] QmS 0.0 0.07 108 42:6 1.0 0.13 85 37:8 2.5 0.35 44 36:6 5.0 1.92 4 34:6 10.0 10.6 0 34:5 IbmS 0.0 0.14 108 43:4 1.0 0.3 84 39:5 2.5 0.8 45 39:1 5.0 4.99 7 38:3 10.0 28.52 0 38:2 Table 6 shows example translations obtained by the three different approaches.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Thus the method makes the fairly strong assumption that the features can be partitioned into two types such that each type alone is sufficient for classification.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"(1), Pr(eI 1) is the language model, which is a trigram language model in this case.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(7), such as the likelihood function used in maximum-entropy problems and other generalized additive models (Lafferty 99).","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In the present study, we limit ourselves to an algorithmic approach, using a deterministic breadthfirst search.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The graph was constructed using 2 million trigrams; we chose these by truncating the parallel datasets up to the number of sentence pairs that contained 2 million trigrams.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"3An English sentence with ambiguous PoS assignment can be trivially represented as a lattice similar to our own, where every pair of consecutive nodes correspond to a word, and every possible PoS assignment for this word is a connecting arc.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We utilized two kinds of datasets in our experiments: (i) monolingual treebanks9 and (ii) large amounts of parallel text with English on one side.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Affix Pron Base category N found N missed (recall) N correct (precision) t,-,7 The second issue is that rare family names can be responsible for overgeneration, especially if these names are otherwise common as single-hanzi words.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The space of a configuration is the sum of the lengths of the nonblank tape contents of the k work tapes.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For irt the Good-Turing estimate just discussed gives us an estimate of p(unseen(f,) I f,)-the probability of observing a previously unseen instance of a construction in ft given that we know that we have a construction in f,.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A formalism generating tree sets with complex path sets can still generate only semilinear languages if its tree sets have independent paths, and semilinear path sets.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"It is chosen such that the decisions it made in including or excluding constituents are most probable under the models for all of the parsers.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"However, we have reason to doubt Chang et al.'s performance claims.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We report results for the best and median hyperparameter settings obtained in this way.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J) and the lexicon probability p(fbi jei) are obtained by relative frequency estimates from the Viterbi alignment path after the final training iteration.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"basically complete, yet some improvements and extensions are still under way.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"It can be shown that this objective is convex in q.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Keys to the table are hashed, using for example Austin Appleby\u2019s MurmurHash2, to integers evenly distributed over a large range.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The difference between the featureless model (+PRIOR) and our full model (+FEATS) is 13.6% and 7.7% average error reduction on best and median settings respectively.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Because b is a function, no additional hypothesis splitting happens.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thus, the tree sets generated by HG's are similar to those of CFG's, with each node annotated by the operation (concatenation or wrapping) used to combine the headed strings derived by the daughters of Tree Adjoining Grammars, a tree rewriting formalism, was introduced by Joshi, Levy and Takahashi (1975) and Joshi (1983\/85).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Section 4 presents experimen tal results on two corpora: the MUC4 terrorism corpus, and Reuters texts about natural disasters.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The Bikel GoldPOS configuration only supplies the gold POS tags; it does not force the parser to use them.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Because we are working with only three parsers, the only situation in which context will help us is when it can indicate we should choose to believe a single parser that disagrees with the majority hypothesis instead of the majority hypothesis itself.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"However, in practice, unknown word models also make the distribution improper.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Reading the following record\u2019s offset indicates where the block ends.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Naseem et al. (2009) and Snyder et al.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"However, we do not explore this possibility in the current work.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The following three sections elaborate these different stages is more detail.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We have not yet tried this.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":", for A. T.&T. nonalpha.. .","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"So, 1: f, xue2shengl+men0 (student+PL) 'students' occurs and we estimate its cost at 11.43; similarly we estimate the cost of f, jiang4+men0 (general+PL) 'generals' (as in 'J' f, xiao3jiang4+men0 'little generals'), at 15.02.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The wounds are still healing.), entity-attribute (e.g., She 2001), who determined that in their corpus of German computer tests, 38% of relations were lexically signalled.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"4.1 Dictionary Representation.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"For instance, on Spanish, the absolute gap on median performance is 10%.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Then, token- level HMM emission parameters are drawn conditioned on these assignments such that each word is only allowed probability mass on a single assigned tag.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In order to solve this problem, a parse tree is needed to understand that \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is not the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We have described two data structures for language modeling that achieve substantial reductions in time and memory cost.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"To improve agreement during the revision process, a dual-blind evaluation was performed in which 10% of the data was annotated by independent teams.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"(c) Coordination ambiguity is shown in dependency scores by e.g., \u00e2\u0088\u0097SSS R) and \u00e2\u0088\u0097NP NP NP R).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In the maximum entropy framework, there is no such constraint.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For E(ni1s), then, we substitute a smooth S against the number of class elements.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Here, an NE instance pair is any pair of NEs separated by at most 4 syntactic chunks; for example, \u00e2\u0080\u009cIBM plans to acquire Lotus\u00e2\u0080\u009d.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Several methods exist to handle collisions; we use linear probing because it has less memory overhead when entries are small.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"(2009) also report results on English, but on the reduced 17 tag set, which is not comparable to ours).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"While the generative power of CG's is greater that of CFG's, it appears to be highly constrained.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Several coreference resolvers have used supervised learning techniques, such as decision trees and rule learners (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Head Grammars (HG's), introduced by Pollard (1984), is a formalism that manipulates headed strings: i.e., strings, one of whose symbols is distinguished as the head.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"3.1 Gross Statistics.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"To make the projection practical, we rely on the twelve universal part-of-speech tags of Petrov et al. (2011).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Our System Wang, Li, and Chang a. 1\\!f!IP Eflltii \/1\\!f!J:P $1til I b. agm: I a m: c. 5 Bf is Bf 1 d. \"*:t: w _t ff 1 \"* :t: w_tff 1 g., , Transliteration\/Translation chen2zhongl-shenl qu3 'music by Chen Zhongshen ' huang2rong2 youlyoul de dao4 'Huang Rong said soberly' zhangl qun2 Zhang Qun xian4zhang3 you2qingl shang4ren2 hou4 'after the county president You Qing had assumed the position' lin2 quan2 'Lin Quan' wang2jian4 'Wang Jian' oulyang2-ke4 'Ouyang Ke' yinl qi2 bu4 ke2neng2 rong2xu3 tai2du2 er2 'because it cannot permit Taiwan Independence so' silfa3-yuan4zhang3 lin2yang2-gang3 'president of the Judicial Yuan, Lin Yanggang' lin2zhangl-hu2 jiangl zuo4 xian4chang3 jie3shuol 'Lin Zhanghu will give an ex\u00c2\u00ad planation live' jin4\/iang3 nian2 nei4 sa3 xia4 de jinlqian2 hui4 ting2zhi3 'in two years the distributed money will stop' gaoltangl da4chi2 ye1zi0 fen3 'chicken stock, a tablespoon of coconut flakes' you2qingl ru4zhu3 xian4fu3 lwu4 'after You Qing headed the county government' Table 5 Performance on morphological analysis.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We realize the importance of paraphrase; however, the major obstacle is the construction of paraphrase knowledge.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For every sequence of initial capitalized words, its longest substring that occurs in the same document as a sequence of initCaps is identified.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In our experiment, we set the threshold of the TF\/ITF score empirically using a small development corpus; a finer adjustment of the threshold could reduce the number of such keywords.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Second, rather than relying on a division of the corpus into manually-assigned portions, we use features intended to capture the usefulness of each phrase pair.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Lemma: If the number of votes required by constituent voting is greater than half of the parsers under consideration the resulting structure has no crossing constituents.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We applied the AutoSlog system (Riloff, 1996) to our unannotated training texts to generate a set of extraction patterns for each domain.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Another way to interpret this is that less than 5% of the correct constituents are missing from the hypotheses generated by the union of the three parsers.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"These sequence models-based approaches commonly treat token-level tag assignment as the primary latent variable.","label":0},{"source_text":"There is no global pruning.","target_text":"Decreasing the threshold results in higher mWER due to additional search errors.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The maximum entropy framework estimates probabilities based on the principle of making as few assumptions as possible, other than the constraints imposed.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This is by now a fairly standard representation for multiple morphological segmentation of Hebrew utterances (Adler, 2001; Bar-Haim et al., 2005; Smith et al., 2005; Cohen and Smith, 2007; Adler, 2007).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Additionally, it works about 3 times as fast as the IBM style search.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The government has to make a decision, and do it quickly.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For a word whose initCaps might be due to its position rather than its meaning (in headlines, first word of a sentence, etc), the case information of other occurrences might be more accurate than its own.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Memory usage is likely much lower than ours. fThe original paper (Germann et al., 2009) provided only 2s of query timing and compared with SRI when it exceeded available RAM.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We define the path set of a tree 1 as the set of strings that label a path from the root to frontier of 7.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"First, the training data for the parser is projectivized by applying a minimal number of lifting operations (Kahane et al., 1998) and encoding information about these lifts in arc labels.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Hyperparameters Our model has two Dirichlet concentration hyperparameters: \u00ce\u00b1 is the shared hyperparameter for the token-level HMM emission and transition distributions.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"where the husband's family name is optionally prepended to the woman's full name; thus ;f:*lf#i xu3lin2-yan2hai3 would represent the name that Ms. Lin Yanhai would take if she married someone named Xu.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Since the parameter and token components will remain fixed throughout experiments, we briefly describe each.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The theory has also been validated empirically.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The remaining arcs are marked OOV.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Tokens tagged as PUNC are not discarded unless they consist entirely of punctuation.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For all variants, we found that BerkeleyLM always rounds the floating-point mantissa to 12 bits then stores indices to unique rounded floats.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"However, the overall percentage of non-projective arcs is less than 2% in PDT and less than 1% in DDT.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"2) An improved language model, which takes into account syntactic structure, e.g. to ensure that a proper English verbgroup is generated.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"This feature includes named entities, which the ATB marks with a flat NP node dominating an arbitrary number of NNP pre-terminal daughters (Figure 2).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"These alignment models are similar to the concept of hidden Markov models (HMM) in speech recognition.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The annotator can then \u00e2\u0080\u009cclick away\u00e2\u0080\u009d those words that are here not used as connectives (such as the conjunction und (\u00e2\u0080\u0098and\u00e2\u0080\u0099) used in lists, or many adverbials that are ambiguous between connective and discourse particle).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"For example, ... fraud related to work on a federally funded sewage plant in Georgia In this case, Georgia is extracted: the NP containing it is a complement to the preposition in; the PP headed by in modifies the NP a federally funded sewage plant, whose head is the singular noun plant.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Note that in our model, conditioned on T , there is precisely one t which has nonzero probability for the token component, since for each word, exactly one \u00ce\u00b8t has support.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The results are given in Table 4.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"All commentaries have been tagged with part-of-speech information using Brants\u00e2\u0080\u0099 TnT1 tagger and the Stuttgart\/Tu\u00c2\u00a8bingen Tag Set automatic analysis was responsible for this decision.)","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If it is made up of all capital letters, then (allCaps, zone) is set to 1.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"A different knowledge source, called CFSemCFSem, compares the semantic expectations of the caseframe that extracts the anaphor with the semantic expectations of the caseframe that extracts the candidate.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Because all English vertices are going to be labeled, we do not need to disambiguate them by embedding them in trigrams.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The confidence intervals are computed by bootstrap resampling for BLEU, and by standard significance testing for the manual scores, as described earlier in the paper.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"This leads to the best reported performance for robust non-projective parsing of Czech.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"A simple lexicalized PCFG with second order Markovization gives relatively poor performance: 75.95% F1 on the test set.8 But this figure is surprisingly competitive with a recent state-of-the-art baseline (Table 7).","label":0},{"source_text":"There is no global pruning.","target_text":"besuchen 9.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We use MMAX for this annotation as well.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Taking only the highest frequency rules is much &quot;safer&quot;, as they tend to be very accurate.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Finally, we incorporate the instance-weighting model into a general linear combination, and learn weights and mixing parameters simultaneously. where c\u03bb(s, t) is a modified count for pair (s, t) in OUT, u(s|t) is a prior distribution, and y is a prior weight.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"These sequence models-based approaches commonly treat token-level tag assignment as the primary latent variable.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For illustration, an English translation of one of the commentaries is given in Figure 1.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"If the same pair of NE instances is used with different phrases, these phrases are likely to be paraphrases.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The details are given in (Och and Ney, 2000).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The BerkeleyLM direct-mapped cache is in principle faster than caches implemented by RandLM and by IRSTLM, so we may write a C++ equivalent implementation as future work.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Table 5: Individual Performance of KSs for Disasters (e.g., \u00e2\u0080\u009cthe mayor\u00e2\u0080\u009d vs. \u00e2\u0080\u009cthe journalist\u00e2\u0080\u009d).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"6One of our experimental settings lacks document boundaries, and we used this approximation in both settings for consistency.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Hyperparameter settings are sorted according to the median one-to-one metric over runs.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For each candidate antecedent, BABAR identifies the caseframe that would extract the candidate, pairs it with the anaphor\u00e2\u0080\u0099s caseframe, and consults the CF Network to see if this pair of caseframes has co-occurred in previous resolutions.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Inflectional features marking pronominal elements may be attached to different kinds of categories marking their pronominal complements.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"mein 5.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For even larger models, storing counts (Talbot and Osborne, 2007; Pauls and Klein, 2011; Guthrie and Hepple, 2010) is a possibility.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"For each language and setting, we report one-to-one (11) and many- to-one (m-1) accuracies.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Each vertex within a connected component must have the same label \u2014 in the binary classification case, we need a single labeled example to identify which component should get which label.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The translation scores for the hypotheses generated with different threshold values t0 are compared to the translation scores obtained with a conservatively large threshold t0 = 10:0 . For each test series, we count the number of sentences whose score is worse than the corresponding score of the test series with the conservatively large threshold t0 = 10:0, and this number is reported as the number of search errors.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Such resources exist for Hebrew (Itai et al., 2006), but unfortunately use a tagging scheme which is incompatible with the one of the Hebrew Treebank.s For this reason, we use a data-driven morphological analyzer derived from the training data similar to (Cohen and Smith, 2007).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"All three models evaluated in this paper incorrectly analyze the constituent as iDafa; none of the models attach the attributive adjectives properly.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This is to allow for fair comparison between the statistical method and GR, which is also purely dictionary-based.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The procedure using the tagged sentences to discover paraphrases takes about one hour on a 2GHz Pentium 4 PC with 1GB of memory.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Proper-Name Identification.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The following recursive equation is evaluated: Qe0 (e; S; C; j) = (2) = p(fj je) max \u00c3\u0086;e00 np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) max (S0;j0) (S0 ;Cnfjg;j0)!(S;C;j) j02Cnfjg Qe00 (e0; S0;C n fjg; j0)o: The search ends in the hypotheses (I; f1; ; Jg; j).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The fraction of buckets that are empty is m\u22121 m , so average lookup time is O( m 1) and, crucially, constant in the number of entries.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We propose an unsupervised method to discover paraphrases from a large untagged corpus, without requiring any seed phrase or other cue.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Se ma nti c (a) filters candidate if its semantic tags d o n \u00e2\u0080\u0099 t i n t e r s e c t w i t h t h o s e o f t h e a n a p h o r .","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Given a sufficient number of randomly drawn unlabeled examples (i.e., edges), we will induce two completely connected components that together span the entire graph.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For the seen word ir, 'gen\u00c2\u00ad erals,' there is an c:NC transduction from to the node preceding ir,; this arc has cost cost( f,) - cost(unseen(f,)), so that the cost of the whole path is the desired cost( f,).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"21 84.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"8 We use head-finding rules specified by a native speaker.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Subsets C of increasing cardinality c are processed.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Each round is composed of two stages; each stage updates one of the classifiers while keeping the other classifier fixed.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Furthermore, we do not connect the English vertices to each other, but only to foreign language vertices.4 The graph vertices are extracted from the different sides of a parallel corpus (De, Df) and an additional unlabeled monolingual foreign corpus Ff, which will be used later for training.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In this example there are four \"input characters,\" A, B, C and D, and these map respectively to four \"pronunciations\" a, b, c and d. Furthermore, there are four \"words\" represented in the dictionary.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We use the universal POS tagset of Petrov et al. (2011) in our experiments.10 This set C consists of the following 12 coarse-grained tags: NOUN (nouns), VERB (verbs), ADJ (adjectives), ADV (adverbs), PRON (pronouns), DET (determiners), ADP (prepositions or postpositions), NUM (numerals), CONJ (conjunctions), PRT (particles), PUNC (punctuation marks) and X (a catch-all for other categories such as abbreviations or foreign words).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The model starts by generating a tag assignment for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Our final average POS tagging accuracy of 83.4% compares very favorably to the average accuracy of Berg-Kirkpatrick et al.\u2019s monolingual unsupervised state-of-the-art model (73.0%), and considerably bridges the gap to fully supervised POS tagging performance (96.6%).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"am 11.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For instance, the common \"suffixes,\" -nia (e.g.,.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Our system does not currently make use of titles, but it would be straightforward to do so within the finite-state framework that we propose.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"When keys are longer than 64 bits, we conserve space by replacing the keys with their 64-bit hashes.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Finally, we intend to explore more sophisticated instanceweighting features for capturing the degree of generality of phrase pairs.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"It can be shown that this objective is convex in q.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Given around 90,000 unlabeled examples, the methods described in this paper classify names with over 91% accuracy.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"2.2.4 Semantic Caseframe Expectations The third type of contextual role knowledge learned by BABAR is Semantic Caseframe Expectations.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models (e.g., (Grosz et al., 1995; Grosz and Sidner, 1998)), syntactic algorithms (e.g., (Hobbs, 1978; Lappin and Le- ass, 1994)), and supervised machine learning systems (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"07 95.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":", December, then the feature MonthName is set to 1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The horizontal axis in this plot represents the most significant dimension, which explains 62% of the variation.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The translation direction is from German to English.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Consider first the examples in (2).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"At the phrasal level, we remove all function tags and traces.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thus, M works in logspace and recognition can be done on a deterministic TM in polynomial tape.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Same for . In the case where the next token is a hyphen, then is also used as a feature: (init- Caps, ) is set to 1.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Formally, let el (62) be the number of classification errors of the first (second) learner on the training data, and let Eco be the number of unlabeled examples on which the two classifiers disagree.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The CoBoost algorithm just described is for the case where there are two labels: for the named entity task there are three labels, and in general it will be useful to generalize the CoBoost algorithm to the multiclass case.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We have argued that the proposed method performs well.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In order to pass these constraints onto the parser, the lexical rules in the grammar are of the form pi \u2014* (si, pi) Parameter Estimation The grammar probabilities are estimated from the corpus using simple relative frequency estimates.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The algorithm, called CoBoost, has the advantage of being more general than the decision-list learning alInput: (xi , yi), , (xim, ) ; x, E 2x, yi = +1 Initialize Di (i) = 1\/m.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(For some recent corpus-based work on Chinese abbreviations, see Huang, Ahrens, and Chen [1993].)","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Using this encoding scheme, the arc from je to Z in Figure 2 would be assigned the label AuxP\u2191Sb (signifying an AuxP that has been lifted from a Sb).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For example, in the CC-domain, 96 keywords are found which have TF\/ITF scores above a threshold; some of them are shown in Figure 3.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"It was then tested on section 22 of the Treebank in conjunction with the other parsers.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"While the first three models get three to four tags wrong, our best model gets only one word wrong and is the most accurate among the four models for this example.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In (1) the sequencema3lu4 cannot be resolved locally, but depends instead upon broader context; similarly in (2), the sequence :::tcai2neng2 cannot be resolved locally: 1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"tai2du2 'Taiwan Independence.'","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Furthermore, some of the arcs represent lexemes not present in the input tokens (e.g. h\/DT, fl\/POS), however these are parts of valid analyses of the token (cf. super-segmental morphology section 2).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The range of the score is between 0 and 1 (higher is better).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We can make several observations on the cause of errors.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Chinese word segmentation can be viewed as a stochastic transduction problem.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"As a result, its POS tag needs to be induced in the \u201cNo LP\u201d case, while the 11A word level paired-t-test is significant at p < 0.01 for Danish, Greek, Italian, Portuguese, Spanish and Swedish, and p < 0.05 for Dutch. correct tag is available as a constraint feature in the \u201cWith LP\u201d case.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The first thing to note is that projectivizing helps in itself, even if no encoding is used, as seen from the fact that the projective baseline outperforms the non-projective training condition by more than half a percentage point on attachment score, although the gain is much smaller with respect to exact match.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Previous work has attempted to incorporate such constraints into token-level models via heavy-handed modifications to inference procedure and objective function (e.g., posterior regularization and ILP decoding) (Grac\u00c2\u00b8a et al., 2009; Ravi and Knight, 2009).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"= p(fj je) max \u00c3\u0086;e00 j02Cnfjg np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j 0 )o: The DP equation is evaluated recursively for each hypothesis (e0; e; C; j).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"As can be seen in Table 4, our training data is a lot less than those used by MENE and IdentiFinder3.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Our code has been publicly available and intergrated into Moses since October 2010.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Our evaluation includes both weighted and un- weighted lattices.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The key to the methods we describe is redundancy in the unlabeled data.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"58 95.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The two knowledge sources that use semantic expectations, WordSemCFSem and CFSemCFSem, always return values of -1 or 0.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The tokens w are generated by token-level tags t from an HMM parameterized by the lexicon structure.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"And indeed, converging on annotation guidelines is even more difficult than it is with co-reference.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A procedural definition to restrict1In the approach described in (Berger et al., 1996), a mor phological analysis is carried out and word morphemes rather than full-form words are used during the search.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Then we ran binary search to determine the least amount of memory with which it would run.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Sima\u2019an et al. (2001) presented parsing results for a DOP tree-gram model using a small data set (500 sentences) and semiautomatic morphological disambiguation.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This token may further embed into a larger utterance, e.g., \u2018bcl hneim\u2019 (literally \u201cin-the-shadow the-pleasant\u201d, meaning roughly \u201cin the pleasant shadow\u201d) in which the dominated Noun is modified by a proceeding space-delimited adjective.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"3.2 Inter-annotator Agreement.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"We use a solution to this problem similar to the one presented in (Och et al., 1999), where target words are joined during training.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We then show that the combining techniques presented above give better parsing accuracy than any of the individual parsers.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"For illustration, an English translation of one of the commentaries is given in Figure 1.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"68 95.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"A large number of rules is needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classi- However, we show that the use of data can reduce the requirements for supervision to just 7 simple &quot;seed&quot; rules.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If it is made up of all capital letters, then (allCaps, zone) is set to 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"For \u00c3\u0086 = 0, no new target word is generated, while an additional source sentence position is covered.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For a language like English, this problem is generally regarded as trivial since words are delimited in English text by whitespace or marks of punctuation.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"5.1 Data Sets.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For natural disasters, BABAR generated 20,479 resolutions: 11,652 from lexical seeding and 8,827 from syntactic seeding.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"For our experiments we also report the mean of precision and recall, which we denote by (P + R)I2 and F-measure.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The goal of machine translation is the translation of a text given in some source language into a target language.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"4.1 Dictionary Representation.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Named Entity Recognition: A Maximum Entropy Approach Using Global Information","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Recently, lattices have been used successfully in the parsing of Hebrew (Tsarfaty, 2006; Cohen and Smith, 2007), a Semitic language with similar properties to Arabic.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We would like to thank Prof. Ralph Grish- man, Mr. Takaaki Hasegawa and Mr. Yusuke Shinyama for useful comments, discussion and evaluation.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Again, the monotone search performs worst.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Taking only the highest frequency rules is much &quot;safer&quot;, as they tend to be very accurate.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Hereafter, each pair of NE categories will be called a domain; e.g. the \u00e2\u0080\u009cCompany \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain, which we will call CC- domain (Step 2).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For natural disasters, BABAR generated 20,479 resolutions: 11,652 from lexical seeding and 8,827 from syntactic seeding.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"This scenario is applicable to a large set of languages and has been considered by a number of authors in the past (Alshawi et al., 2000; Xi and Hwa, 2005; Ganchev et al., 2009).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"It can also be seen clearly in this plot that two of the Taiwan speakers cluster very closely together, and the third Tai\u00c2\u00ad wan speaker is also close in the most significant dimension (the x axis).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Table 4 Differences in performance between our system and Wang, Li, and Chang (1992).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Finally, we make some improvements to baseline approaches.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Test set OOV rate is computed using the following splits: ATB (Chiang et al., 2006); CTB6 (Huang and Harper, 2009); Negra (Dubey and Keller, 2003); English, sections 221 (train) and section 23 (test).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Of course, we are primarily interested in applying our techniques to languages for which no labeled resources are available.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Using the concept of inverted alignments, we explicitly take care of the coverage constraint by introducing a coverage set C of source sentence positions that have been already processed.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Their theoretical finding is simply stated: classification error rate decreases toward the noise rate exponentially in the number of independent, accurate classifiers.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"0 57.2 43.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In this work we extended the AdaBoost.MH (Schapire and Singer 98) algorithm to the cotraining case.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The sentence-selection approach is crude in that it imposes a binary distinction between useful and non-useful parts of OUT.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"In our particular case the majority requires the agreement of only two parsers because we have only three.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"we perform five runs with different random initialization of sampling state.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The translation of one position in the source sentence may be postponed for up to L = 3 source positions, and the translation of up to two source positions may be anticipated for at most R = 10 source positions.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Tsarfaty (2006) was the first to demonstrate that fully automatic Hebrew parsing is feasible using the newly available 5000 sentences treebank.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This section describes an algorithm based on boosting algorithms, which were previously developed for supervised machine learning problems.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"0 Figure 5 An example of affixation: the plural affix.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Since we could not bias the subjects towards a particular segmentation and did not presume linguistic sophistication on their part, the instructions were simple: subjects were to mark all places they might plausibly pause if they were reading the text aloud.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We can 5 Recall that precision is defined to be the number of correct hits divided by the total number of items.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If all three sentences are in the same document, then even a human will find it difficult to classify McCann in (1) into either person or organization, unless there is some other information provided.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This makes memory usage comparable to our PROBING model.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"While building a machine translation system is a serious undertaking, in future we hope to attract more newcomers to the field by keeping the barrier of entry as low as possible.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We tagged each noun with the top-level semantic classes assigned to it in Word- Net.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"All notions of word, with the exception of the orthographic word, are as relevant in Chinese as they are in English, and just as is the case in other languages, a word in Chinese may correspond to one or more symbols in the orthog 1 For a related approach to the problem of word-segrnention in Japanese, see Nagata (1994), inter alia..","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We utilized two kinds of datasets in our experiments: (i) monolingual treebanks9 and (ii) large amounts of parallel text with English on one side.","label":0},{"source_text":"There is no global pruning.","target_text":"The complexity of the algorithm is O(E3 J2 2J), where E is the size of the target language vocabulary.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Another question that remains unanswered is to what extent the linguistic information he considers can be handled-or at least approximated-by finite-state language models, and therefore could be directly interfaced with the segmentation model that we have presented in this paper.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In addition we report for each model its performance on goldsegmented input (GS) to indicate the upper bound 11Overt definiteness errors may be seen as a wrong feature rather than as wrong constituent and it is by now an accepted standard to report accuracy with and without such errors. for the grammars\u2019 performance on the parsing task.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"None of the models we have presented utilize features associated with a particular constituent (i.e. the label, span, parent label, etc.) to influence parser preference.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"both MENE and IdentiFinder used more training data than we did (we used only the official MUC 6 and MUC7 training data).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"However, their system is a hybrid of hand-coded rules and machine learning methods.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Subsets C of increasing cardinality c are processed.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For each token , zero, one, or more of the features in each feature group are set to 1.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The task is to learn a function from an input string (proper name) to its type, which we will assume to be one of the categories Person, Organization, or Location.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The model was parameterized such that the joint probability of a (label, feature-set) pair P(yi, xi) is written as The model assumes that (y, x) pairs are generated by an underlying process where the label is first chosen with some prior probability P(yi); the number of features mi is then chosen with some probability P(mi); finally the features are independently generated with probabilities P(xulyi).","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"3.4 Salience-based text generation.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"It is important to bear in mind, though, that this is not an inherent limitation of the model.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Formally, for a lexicon L and segments I \u00e2\u0088\u0088 L, O \u00e2\u0088\u0088\/ L, each word automaton accepts the language I\u00e2\u0088\u0097(O + I)I\u00e2\u0088\u0097.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Maamouri et al.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The Bayes models were able to achieve significantly higher precision than their non-parametric counterparts.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"This limits the number of NE category pairs to 2,000 and the number of NE pair instances to 0.63 million.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Two annotators received training with the RST definitions and started the process with a first set of 10 texts, the results of which were intensively discussed and revised.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The iw all map variant uses a non-0 y weight on a uniform prior in p,,(s t), and outperforms a version with y = 0 (iw all) and the \u201cflattened\u201d variant described in section 3.2.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In (b) is a plausible segmentation for this sentence; in (c) is an implausible segmentation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"An example is in (i), where the system fails to group t;,f;?\"$?t!: lin2yang2gang3 as a name, because all three hanzi can in principle be separate words (t;,f; lin2 'wood';?\"$ yang2 'ocean'; ?t!; gang3 'harbor').","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"It is worth noting that, although nonprojective constructions are less frequent in DDT than in PDT, they seem to be more deeply nested, since only about 80% can be projectivized with a single lift, while almost 95% of the non-projective arcs in PDT only require a single lift.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"For example, in the CC-domain, 96 keywords are found which have TF\/ITF scores above a threshold; some of them are shown in Figure 3.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Standard SMT systems have a hierarchical parameter structure: top-level log-linear weights are used to combine a small set of complex features, interpreted as log probabilities, many of which have their own internal parameters and objectives.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We call these N \u2212 1 words state.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Our empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Unigram lookup is dense so we use an array of probability and backoff values.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"7).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"orthographic words are thus only a starting point for further analysis and can only be regarded as a useful hint at the desired division of the sentence into words.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This is especially effective at reducing load time, since raw bytes are read directly to memory\u2014or, as happens with repeatedly used models, are already in the disk cache.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE is then trained on 80% of the training corpus, and tested on the remaining 20%.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"An initial step of any text\u00c2\u00ad analysis task is the tokenization of the input into words.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The development of a na\u00efve Bayes classifier involves learning how much each parser should be trusted for the decisions it makes.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"An ATM may be thought of as spawning independent processes for each applicable move.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Recent research has demonstrated that incorporating this sparsity constraint improves tagging accuracy.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The 14 general-language features embody straightforward cues: frequency, \u201ccentrality\u201d as reflected in model scores, and lack of burstiness.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Interpolation search formalizes the notion that one opens a dictionary near the end to find the word \u201czebra.\u201d Initially, the algorithm knows the array begins at b +\u2014 0 and ends at e +\u2014 |A|\u22121.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Here we use a slightly different notion of lift, applying to individual arcs and moving their head upwards one step at a time: Intuitively, lifting an arc makes the word wk dependent on the head wi of its original head wj (which is unique in a well-formed dependency graph), unless wj is a root in which case the operation is undefined (but then wj \u2014* wk is necessarily projective if the dependency graph is well-formed).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In (Reitter, Stede 2003) we went a different way and suggested URML5, an XML format for underspecifying rhetorical structure: a number of relations can be assigned instead of a single one, competing analyses can be represented with shared forests.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Caching for IRSTLM is smaller at 0.09 GB resident memory, though it supports only a single thread.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For example, in the sentence that starts with \u00e2\u0080\u009cBush put a freeze on . . .","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"keywords Step 3 Sets of phrases based on keywords Step 4 Links between sets of phrases All the contexts collected for a given domain are gathered in a bag and the TF\/ITF scores are calculated for all the words except stopwords in the bag.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The problem can be represented as a graph with 2N vertices corresponding to the members of X1 and X2.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"This departure from the traditional token-based tagging approach allows us to explicitly capture type- level distributional properties of valid POS tag as signments as part of the model.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Juri Ganitkevitch answered questions about Joshua.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"This approach leads to a search procedure with complexity O(E3 J4).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(e.g., N.Y. would contribute this feature, IBM would not). nonalpha=x Appears if the spelling contains any characters other than upper or lower case letters.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Figure 1 shows an example.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"However, it is possible to personify any noun, so in children's stories or fables, i\u00c2\u00a5JJ1l.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Learned Tag Prior (PRIOR) We next assume there exists a single prior distribution \u00cf\u0088 over tag assignments drawn from DIRICHLET(\u00ce\u00b2, K ).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"BABAR applies a DempsterShafer probabilistic model to make resolutions based on evidence from the contextual role knowledge sources as well as general knowledge sources.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Our experiments consistently demonstrate that this model architecture yields substantial performance gains over more complex tagging counterparts.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This PP modifies another NP, whose head is a singular noun.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Hyperparameters Our model has two Dirichlet concentration hyperparameters: \u00ce\u00b1 is the shared hyperparameter for the token-level HMM emission and transition distributions.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We can 5 Recall that precision is defined to be the number of correct hits divided by the total number of items.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The compact variant uses sorted arrays instead of hash tables within each node, saving some memory, but still stores full 64-bit pointers.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"As expected, the vanilla HMM trained with EM performs the worst.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This heuristics is used to prune all segmentation possibilities involving \u201clexically improper\u201d segments.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Thus, we feel fairly confident that for the examples we have considered from Gan's study a solution can be incorporated, or at least approximated, within a finite-state framework.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We simulate lexical constraints by using an external lexical resource against which we verify whether OOV segments are in fact valid Hebrew lexemes.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"a classifier.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"pronunciation depends upon word affiliation: tfJ is pronounced deO when it is a prenominal modification marker, but di4 in the word \u00c2\u00a7tfJ mu4di4 'goal'; fl; is normally ganl 'dry,' but qian2 in a person's given name.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Standard SMT systems have a hierarchical parameter structure: top-level log-linear weights are used to combine a small set of complex features, interpreted as log probabilities, many of which have their own internal parameters and objectives.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"This work was supported in part by the Swedish Research Council (621-2002-4207).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This information is readily available in TRIE where adjacent records with equal pointers indicate no further extension of context is possible.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Acknowledgments We thank Steven Bethard, Evan Rosen, and Karen Shiells for material contributions to this work.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We evaluate the time and memory consumption of each data structure by computing perplexity on 4 billion tokens from the English Gigaword corpus (Parker et al., 2009).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"It is difficult to motivate these days why one ministry should be exempt from cutbacks \u00e2\u0080\u0094 at the expense of the others.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"A contextual role represents the role that a noun phrase plays in an event or relationship.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The local features used are similar to those used in BBN' s IdentiFinder (Bikel et al., 1999) or MENE (Borthwick, 1999).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Note that in our model, conditioned on T , there is precisely one t which has nonzero probability for the token component, since for each word, exactly one \u00ce\u00b8t has support.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"It is difficult to directly compare the Matsoukas et al results with ours, since our out-of-domain corpus is homogeneous; given heterogeneous training data, however, it would be trivial to include Matsoukas-style identity features in our instance-weighting model.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, there are several reasons why this approach will not in general work: 1.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Somewhat surprisingly, there do not appear to be large systematic differences between linear and MAP combinations.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(1999) did not report using any dictionaries, but mentioned in a footnote that they have added list membership features, which have helped marginally in certain domains.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We then gather all phrases with the same keyword.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Using less training data than other systems, our NER is able to perform as well as other state-of-the-art NERs.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"A dynamic programming recursion similar to the one in Eq. 2 is evaluated.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"(2009).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A morpheme, on the other hand, usually corresponds to a unique hanzi, though there are a few cases where variant forms are found.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Another question that remains unanswered is to what extent the linguistic information he considers can be handled-or at least approximated-by finite-state language models, and therefore could be directly interfaced with the segmentation model that we have presented in this paper.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"On the other hand, if it is seen as McCann Pte.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"In MSA, SVO usually appears in non-matrix clauses.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The first value reports resident size after loading; the second is the gap between post-loading resident memory and peak virtual memory.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, suppose the current model assigns a belief value of .60 to {A, B}, meaning that it is 60% sure that the correct hypothesis is either A or B. Then new evidence arrives with a belief value of .70 assigned 5 Initially there are no competing hypotheses because all hypotheses are included in \u00ce\u00b8 by definition.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"There are two differences between this method and the DL-CoTrain algorithm: spelling and contextual features, alternating between labeling and learning with the two types of features.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"All the sentences have been analyzed by our chunker and NE tag- ger.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The first point we need to address is what type of linguistic object a hanzi repre\u00c2\u00ad sents.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Unification is used in LFG's to produce structures having two dependent spines of unbounded length as in Figure 5.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The word joining is done on the basis of a likelihood criterion.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"00 76.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Table 4 shows the results.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Our full model (\u201cWith LP\u201d) outperforms the unsupervised baselines and the \u201cNo LP\u201d setting for all languages.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We performed three experiments to evaluate our techniques.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We compare the manually annotated grammar, which we incorporate into the Stanford parser, to both the Berkeley (Petrov et al., 2006) and Bikel (Bikel, 2004) parsers.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Here NO counts the number of hypothesized constituents in the development set that match the binary predicate specified as an argument.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This differs from other implementations (Stolcke, 2002; Pauls and Klein, 2011) that use hash tables as nodes in a trie, as explained in the next section.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Given around 90,000 unlabeled examples, the methods described in this paper classify names with over 91% accuracy.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"RandLM is the clear winner in RAM utilization, but is also slower and lower quality.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Once again we present both a non-parametric and a parametric technique for this task.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The samples from each corpus were independently evaluated.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Find keywords for each NE pair When we look at the contexts for each domain, we noticed that there is one or a few important words which indicate the relation between the NEs (for example, the word \u00e2\u0080\u009cunit\u00e2\u0080\u009d for the phrase \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Recent results (e.g., (Yarowsky 95; Brill 95; Blum and Mitchell 98)) have suggested that unlabeled data can be used quite profitably in reducing the need for supervision.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Others depend upon various lexical heuris\u00c2\u00ad tics: for example Chen and Liu (1992) attempt to balance the length of words in a three-word window, favoring segmentations that give approximately equal length for each word.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"However, it is unlikely that other occurrences of News Broadcasting Corp. in the same document also co-occur with Even.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Test set OOV rate is computed using the following splits: ATB (Chiang et al., 2006); CTB6 (Huang and Harper, 2009); Negra (Dubey and Keller, 2003); English, sections 221 (train) and section 23 (test).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"4.2 Global Features.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"A name already mentioned previously in a document may appear in abbreviated form when it is mentioned again later.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Along with IRSTLM and TPT, our binary format is memory mapped, meaning the file and in-memory representation are the same.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The features are used to represent each example for the learning algorithm.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"TPT has theoretically better locality because it stores ngrams near their suffixes, thereby placing reads for a single query in the same or adjacent pages.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The maximum precision oracle is an upper bound on the possible gain we can achieve by parse hybridization.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This is not ideal for some applications, however.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The local features used are similar to those used in BBN' s IdentiFinder (Bikel et al., 1999) or MENE (Borthwick, 1999).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Lexicalizing several POS tags improves performance.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Figure 2 shows a screenshot (which is of somewhat limited value, though, as color plays a major role in signalling the different statuses of the information).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Except for the left3The graphs satisfy all the well-formedness conditions given in section 2 except (possibly) connectedness.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In contrast, even though context can be helpful for resolving definite NPs, context can be trumped by the semantics of the nouns themselves.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This suggests a direct parallel to (1): where \u02dcp(s, t) is a joint empirical distribution extracted from the IN dev set using the standard procedure.2 An alternative form of linear combination is a maximum a posteriori (MAP) combination (Bacchiani et al., 2004).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"For example, the formalisms in the hierarchy described above generate semilinear languages although their path sets become increasingly more complex as one moves up the hierarchy.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"21 84.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"By applying an inverse transformation to the output of the parser, arcs with non-standard labels can be lowered to their proper place in the dependency graph, giving rise 1The dependency graph has been modified to make the final period a dependent of the main verb instead of being a dependent of a special root node for the sentence. to non-projective structures.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We show the results of three of the experiments we conducted to measure isolated constituent precision under various partitioning schemes.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The second concerns the methods used (if any) to ex\u00c2\u00ad tend the lexicon beyond the static list of entries provided by the machine-readable dictionary upon which it is based.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We propose a limit of 70 words for Arabic parsing evaluations.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"In both cases, the instanceweighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline, and gains of between 0.6 and 1.8 over an equivalent mixture model (with an identical training procedure but without instance weighting).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This has the effect of randomly permuting vocabulary identifiers, meeting the requirements of interpolation search when vocabulary identifiers are used as keys.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"For each language, we took the same number of sentences from the bitext as there are in its treebank, and trained a supervised feature-HMM.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"4.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Despite these limitations, a purely finite-state approach to Chinese word segmentation enjoys a number of strong advantages.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"(c) Coordination ambiguity is shown in dependency scores by e.g., \u00e2\u0088\u0097SSS R) and \u00e2\u0088\u0097NP NP NP R).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"(1998) did make use of information from the whole document.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Therefore, the number of fine tags varied across languages for our experiments; however, one could as well have fixed the set of HMM states to be a constant across languages, and created one mapping to the universal POS tagset.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"10 Other orthographic normalization schemes have been suggested for Arabic (Habash and Sadat, 2006), but we observe negligible parsing performance differences between these and the simple scheme used in this evaluation.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The Grammar Our parser looks for the most likely tree spanning a single path through the lattice of which the yield is a sequence of lexemes.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"This work was supported in part by the Swedish Research Council (621-2002-4207).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For verbs we add two features.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Set the decision list to include all rules whose (smoothed) strength is above some threshold Pmin.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"On the surface, our model may seem as a special case of Cohen and Smith in which \u03b1 = 0.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Such constraints are derived from training data, expressing some relationship between features and outcome.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"summit Sharm (a) Al-Sheikh summit Sharm (b) DTNNP Al-Sheikh in a corpus position without a bracketing label, then we also add \u00e2\u0088\u0097n, NIL) to M. We call the set of unique n-grams with multiple labels in M the variation nuclei of C. Bracketing variation can result from either annotation errors or linguistic ambiguity.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"While processing the source sentence monotonically, the initial state I is entered whenever there are no uncovered positions to the left of the rightmost covered position.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"At each point during the derivation, the prediction is based on six word tokens, the two topmost tokens on the stack, and the next four input tokens.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"This research was supported in part by the Defense Advanced Research Projects Agency as part of the Translingual Information Detection, Extraction and Summarization (TIDES) program, under Grant N66001001-18917 from the Space and Naval Warfare Systems Center, San Diego, and by the National Science Foundation under Grant IIS00325657.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"keywords Step 3 Sets of phrases based on keywords Step 4 Links between sets of phrases All the contexts collected for a given domain are gathered in a bag and the TF\/ITF scores are calculated for all the words except stopwords in the bag.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":".","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This combination generalizes (2) and (3): we use either at = a to obtain a fixed-weight linear combination, or at = cI(t)\/(cI(t) + 0) to obtain a MAP combination.","label":0},{"source_text":"Here we present two algorithms.","target_text":"An important reason for separating the two types of features is that this opens up the possibility of theoretical analysis of the use of unlabeled examples.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The breakdown of the different types of words found by ST in the test corpus is given in Table 3.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"02 99.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The correct resolution in sentence (c) depends on knowledge that kidnappers frequently blindfold their victims.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"However, it can be noted that the results for the least informative encoding, Path, are almost comparable, while the third encoding, Head, gives substantially worse results for both data sets.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The Hebrew token \u2018bcl\u20191, for example, stands for the complete prepositional phrase 'We adopt here the transliteration of (Sima\u2019an et al., 2001).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"A different but supplementary perspective on discourse-based information structure is taken 11ventionalized patterns (e.g., order of informa by one of our partner projects, which is inter tion in news reports).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The 14 general-language features embody straightforward cues: frequency, \u201ccentrality\u201d as reflected in model scores, and lack of burstiness.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Entries for 2 < n < N store a vocabulary identifier, probability, backoff, and an index into the n + 1-gram table.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Lexicon Feature of Previous and Next Token: The string of the previous token and the next token is used with the initCaps information of . If has initCaps, then a feature (initCaps, ) is set to 1.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We first give a brief overview of boosting algorithms.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This paper describes the several performance techniques used and presents benchmarks against alternative implementations.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Precision is the portion of hypothesized constituents that are correct and recall is the portion of the Treebank constituents that are hypothesized.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"5 Related Work.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The German finite verbs 'bin' (second example) and 'konnten' (third example) are too far away from the personal pronouns 'ich' and 'Sie' (6 respectively 5 source sentence positions).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"63 95.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"First, we use a novel graph-based framework for projecting syntactic information across language boundaries.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The web-based Annis imports data in a variety of XML formats and tagsets and displays it in a tier-orientedway (optionally, trees can be drawn more ele gantly in a separate window).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The price shocked her.), or same-kind (e.g., Her health insurance paid for the hospital fees, but the automobile insurance did not cover the repair.).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"6 Joint Segmentation and Parsing.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This is a straightforward technique that is arguably better suited to the adaptation task than the standard method of treating representative IN sentences as queries, then pooling the match results.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This is especially bad with PROBING because it is based on hashing and performs random lookups, but it is not intended to be used in low-memory scenarios.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The parser builds dependency graphs by traversing the input from left to right, using a stack to store tokens that are not yet complete with respect to their dependents.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"For the English RST-annotated corpus that is made available via LDC, his corresponding result is 62%.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We refer to this process as Reliable Case Resolution because it involves finding cases of anaphora that can be easily resolved with their antecedents.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The content does not necessarily reflect the views of the U.S. Government, and no official endorsement should be inferred.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For the disasters domain, 8245 texts were used for training and the 40 test documents contained 447 anaphoric links.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"This withdrawal by the treasury secretary is understandable, though.","label":1},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The corpus has been annotated with six different types of information, which are characterized in the following subsections.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Then, moving from connective to connective, ConAno sometimes offers suggestions for its scope (using heuristics like \u00e2\u0080\u0098for sub- junctor, mark all words up to the next comma as the first segment\u00e2\u0080\u0099), which the annotator can accept with a mouseclick or overwrite, marking instead the correct scope with the mouse.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Here a set is represented by the keyword and the number in parentheses indicates the number of shared NE pair instances.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We therefore used the arithmetic mean of each interjudge precision-recall pair as a single measure of interjudge similarity.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In Eq.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Two processes are spawned requiring B to derive z,.., and C to derive yi , , y,.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The annotator can then \u00e2\u0080\u009cclick away\u00e2\u0080\u009d those words that are here not used as connectives (such as the conjunction und (\u00e2\u0080\u0098and\u00e2\u0080\u0099) used in lists, or many adverbials that are ambiguous between connective and discourse particle).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"This problem arises because our keywords consist of only one word.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In this section, we describe the seven general knowledge sources and explain how the DempsterShafer model makes resolutions.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In various dialects of Mandarin certain phonetic rules apply at the word.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"handled given appropriate models.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"They are set to fixed constants.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"While we cannot prove there are no such useful features on which one should condition trust, we can give some insight into why the features we explored offered no gain.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For MUC7, there are also no published results on systems trained on only the official training data of 200 aviation disaster articles.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The tool we use is MMAX8, which has been specifically designed for marking co-reference.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"This group of features attempts to capture such information.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"yu2 'fish.'","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Our parsing performance measures (SY N) thus report the PARSEVAL extension proposed in Tsarfaty (2006).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In our experiments we set the parameter values randomly, and then ran EM to convergence.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Our baseline for all sentence lengths is 5.23% F1 higher than the best previous result.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In Table 7 we give results for several evaluation metrics.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"First of all, most previous articles report perfor\u00c2\u00ad mance in terms of a single percent-correct score, or else in terms of the paired measures of precision and recall.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Extract NE pair instances with contexts From the four years of newspaper corpus, we extracted 1.9 million pairs of NE instances.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"7 68.3 56.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Much could be done to further reduce memory consumption.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Even when there is training data available in the domain of interest, there is often additional data from other domains that could in principle be used to improve performance.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Both parametric and non-parametric models are explored.","label":1},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The AdaBoost algorithm was developed for supervised learning.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Moreover, they are used as substantives much 2 Unlike machine translation, constituency parsing is not significantly affected by variable word order.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The accuracy is calculated as the ratio of the number of paraphrases to the total number of phrases in the set.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The pseudo-code describing the algorithm is given in Fig.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"However, when grammatical relations like subject and object are evaluated, parsing performance drops considerably (Green et al., 2009).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While building a machine translation system is a serious undertaking, in future we hope to attract more newcomers to the field by keeping the barrier of entry as low as possible.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The second case involves existential noun phrases (Allen, 1995), which are noun phrases that uniquely specify an object or concept and therefore do not need a prior referent in the discourse.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This assumption, however, is not inherent to type-based tagging models.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The cost of storing these averages, in bits, is Because there are comparatively few unigrams, we elected to store them byte-aligned and unquantized, making every query faster.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"5 \u00e2\u0080\u0098Underspecified Rhetorical Markup Language\u00e2\u0080\u0099 6 This confirms the figure given by (Schauer, Hahn.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kinds of cue.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Note that Zt is a normalization constant that ensures the distribution Dt+i sums to 1; it is a function of the weak hypothesis ht and the weight for that hypothesis at chosen at the tth round.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Our analysis identifies three key factors driving our performance gain: 1) selecting a model structure which directly encodes tag sparsity, 2) a type-level prior on tag assignments, and 3) a straightforward na\u00c2\u00a8\u00c4\u00b1veBayes approach to incorporate features.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"In graph-based learning approaches one constructs a graph whose vertices are labeled and unlabeled examples, and whose weighted edges encode the degree to which the examples they link have the same label (Zhu et al., 2003).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"11 taTweel (-) is an elongation character used in Arabic script to justify text.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We define a symmetric similarity function K(uZ7 uj) over two foreign language vertices uZ7 uj E Vf based on the co-occurrence statistics of the nine feature concepts given in Table 1.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Applying the function PROJECTIVIZE to the graph in Figure 1 yields the graph in Figure 2, where the problematic arc pointing to Z has been lifted from the original head jedna to the ancestor je.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"In both cases, SRILM walks its trie an additional time to minimize context as mentioned in Section 4.1.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"To check for this, we do pairwise bootstrap resampling: Again, we repeatedly sample sets of sentences, this time from both systems, and compare their BLEU scores on these sets.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Tsarfaty (2006) argues that for Semitic languages determining the correct morphological segmentation is dependent on syntactic context and shows that increasing information sharing between the morphological and the syntactic components leads to improved performance on the joint task.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Our experiments consistently demonstrate that this model architecture yields substantial performance gains over more complex tagging counterparts.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Nodes in the trie are based on arrays sorted by vocabulary identifier.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"One of the strengths of the DempsterShafer model is its natural ability to recognize when several credible hypotheses are still in play.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Clearly, for judges h and h taking h as standard and computing the precision and recall for Jz yields the same results as taking h as the standard, and computing for h, 14 All evaluation materials, with the exception of those used for evaluating personal names were drawn.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"There are still some open issues to be resolved with the format, but it represents a first step.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Table 4 shows the results.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For a trigram language model, the partial hypotheses are of the form (e0; e; C; j).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Precision.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Thus we are interested not in extraction, but actual generation from representations that may be developed to different degrees of granularity.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"There is no global pruning.","target_text":"We apply a beam search concept as in speech recognition.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"More recently, Subramanya et al. (2010) defined a graph over the cliques in an underlying structured prediction model.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Some relations are signalled by subordinating conjunctions, which clearly demarcate the range of the text spans related (matrix clause, embedded clause).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"(2010)\u00e2\u0080\u0099s richest model: optimized via either EM or LBFGS, as their relative performance depends on the language.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Sometimes extracted phrases by themselves are not meaningful to consider without context, but we set the following criteria.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Reiche\u00e2\u0080\u0099s colleagues will make sure that the concept is waterproof.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"By contrast, BerkeleyLM\u2019s hash and compressed variants will return incorrect results based on an n \u22121-gram.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Lilo CFG's, TAG's, and HG's the derivation tree set of a MCTAG will be a local set.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"pronunciation depends upon word affiliation: tfJ is pronounced deO when it is a prenominal modification marker, but di4 in the word \u00c2\u00a7tfJ mu4di4 'goal'; fl; is normally ganl 'dry,' but qian2 in a person's given name.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"It is not immediately obvious how to formulate an equivalent to equation (1) for an adapted TM, because there is no well-defined objective for learning TMs from parallel corpora.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In this paper k = 3 (the three labels are person, organization, location), and we set a = 0.1.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"2.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Independence of paths at this level reflects context freeness of rewriting and suggests why they can be recognized efficiently.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Following the system devised under the Qing emperor Kang Xi, hanzi have traditionally been classified according to a set of approximately 200 semantic radicals; members of a radical class share a particular structural component, and often also share a common meaning (hence the term 'semantic').","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The techniques we develop can be extended in a relatively straightforward manner to the more general case when OUT consists of multiple sub-domains.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Our analysis and comparison focuses primarily on the one-to-one accuracy since it is a stricter metric than many-to-one accuracy, but also report many-to-one for completeness.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"However, we do not explore this possibility in the current work.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We have checked if there are similar verbs in other major domains, but this was the only one.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"For this experiment, we compare our model with the uniform tag assignment prior (1TW) with the learned prior (+PRIOR).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Clearly, retaining the original frequencies is important for good performance, and globally smoothing the final weighted frequencies is crucial.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"On average, 6 reference translations per automatic translation are available.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Within this framework, we use features intended to capture degree of generality, including the output from an SVM classifier that uses the intersection between IN and OUT as positive examples.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Fortunately, we were able to obtain a copy of the full set of sentences from Chang et al. on which Wang, Li, and Chang tested their system, along with the output of their system.18 In what follows we will discuss all cases from this set where our performance on names differs from that of Wang, Li, and Chang.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"2.1 Inverted Alignments.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"For a corpus C, let M be the set of tuples \u00e2\u0088\u0097n, l), where n is an n-gram with bracketing label l. If any n appears 6 Generative parsing performance is known to deteriorate with sentence length.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"\u00e2\u0080\u009cAgree\u00e2\u0080\u009d is a subject control verb, which dominates another verb whose subject is the same as that of \u00e2\u0080\u009cagree\u00e2\u0080\u009d; the latter verb is generally the one of interest for extraction.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Since our destructor is an efficient call to munmap, bypassing the destructor favors only other packages.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"6 Results and Analysis.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"By making use of global context, it has achieved excellent results on both MUC6 and MUC7 official test data.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This paper describes the several performance techniques used and presents benchmarks against alternative implementations.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"A large number of rules is needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classi- However, we show that the use of data can reduce the requirements for supervision to just 7 simple &quot;seed&quot; rules.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As a partial solution, for pairs of hanzi that co-occur sufficiently often in our namelists, we use the estimated bigram cost, rather than the independence-based cost.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"3 61.7 38.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The approach uses both spelling and contextual rules.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"\u00e2\u0088\u0097NP NP PP R) and \u00e2\u0088\u0097NP NP ADJP R) are both iDafa attachment.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each domain, we created a blind test set by manually annotating 40 doc uments with anaphoric chains, which represent sets of m3 (S) = ) X \u00e2\u0088\u00a9Y =S 1 \u00e2\u0088\u0092 ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) (1) noun phrases that are coreferent (as done for MUC6 (MUC6 Proceedings, 1995)).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The simplest version of the maximum matching algorithm effectively deals with ambiguity by ignoring it, since the method is guaranteed to produce only one segmentation.","label":0},{"source_text":"There is no global pruning.","target_text":"Table 3: Training and test conditions for the Verbmobil task (*number of words without punctuation marks).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"1,000 of these were picked at random, and labeled by hand to produce a test set.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"But we also need an estimate of the probability for a non-occurring though possible plural form like i\u00c2\u00a5JJ1l.f, nan2gua1-men0 'pumpkins.'","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Since these path sets are semilinear, the property of independent paths in their tree sets is sufficient to cause semilinearity of the languages generated by them.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Thus at each iteration the method induces at most n x k rules, where k is the number of possible labels (k = 3 in the experiments in this paper). step 3.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"At the same time, the n-gram error rate is sensitive to samples with extreme n-gram counts.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Along with locking and background kernel operations such as prefaulting, this explains why wall time is not one-eighth that of the single-threaded case. aLossy compression with the same weights. bLossy compression with retuned weights. the non-lossy options.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"As can be seen in Table 4, our training data is a lot less than those used by MENE and IdentiFinder3.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"To approximate these baselines, we implemented a very simple sentence selection algorithm in which parallel sentence pairs from OUT are ranked by the perplexity of their target half according to the IN language model.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"ICOC and CSPP contributed the greatest im provements.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"On the other hand, given that all schemes have similar parsing accuracy overall, this means that the Path scheme is the least likely to introduce errors on projective arcs.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"A cell in the bottom row of the parse chart is required for each potential whitespace boundary.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"This is by now a fairly standard representation for multiple morphological segmentation of Hebrew utterances (Adler, 2001; Bar-Haim et al., 2005; Smith et al., 2005; Cohen and Smith, 2007; Adler, 2007).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Based on revision 4041, we modified Moses to print process statistics before terminating.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Each word is simply tagged with the semantic classes corresponding to all of its senses.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The effect of a second reference resolution classifier is not entirely the same as that of global features.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Memory usage in PROBING is high, though SRILM is even larger, so where memory is of concern we recommend using TRIE, if it fits in memory.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The attachment in such cases encompasses a long distance dependency that cannot be captured by Markovian processes that are typically used for morphological disambiguation.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We consider two variants of Berg-Kirkpatrick et al.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Another attempt at using global information can be found in (Borthwick, 1999).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"If there are too many distinct states, the decoder prunes low-scoring partial hypotheses, possibly leading to a search error.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"kann 7.nicht 8.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The goal of machine translation is the translation of a text given in some source language into a target language.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We of course also fail to identify, by the methods just described, given names used without their associated family name.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"\u00c2\u00a3 : _ADV: 5.88 If:!","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We have argued that the proposed method performs well.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For each source word f, the list of its possible translations e is sorted according to p(fje) puni(e), where puni(e) is the unigram probability of the English word e. It is su\u00c3\u0086cient to consider only the best 50 words.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Thus in a two-hanzi word like lflli?J zhong1guo2 (middle country) 'China' there are two syllables, and at the same time two morphemes.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"To facilitate comparison with previous work, we exhaustively evaluate this grammar and two other parsing models when gold segmentation is assumed (\u00c2\u00a75).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The first thing to note is that projectivizing helps in itself, even if no encoding is used, as seen from the fact that the projective baseline outperforms the non-projective training condition by more than half a percentage point on attachment score, although the gain is much smaller with respect to exact match.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"na me =>1 ha nzi fa mi ly 2 ha nzi gi ve n 3.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Note that because we extracted only high-confidence alignments, many foreign vertices will not be connected to any English vertices.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This is an iterative method that improves the estimation of the parameters at each iteration.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our coreference resolver performed well in two domains, and experiments showed that each contextual role knowledge source contributed valuable information.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"This feature includes named entities, which the ATB marks with a flat NP node dominating an arbitrary number of NNP pre-terminal daughters (Figure 2).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Note that in line 4 the last visited position for the successor hypothesis must be m. Otherwise , there will be four uncovered positions for the predecessor hypothesis violating the restriction.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The Wang, Li, and Chang system fails on fragment (b) because their system lacks the word youlyoul 'soberly' and misinterpreted the thus isolated first youl as being the final hanzi of the preceding name; similarly our system failed in fragment (h) since it is missing the abbreviation i:lJI!","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"This remains as future work.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Following Sproat and Shih (1990), performance for Chinese segmentation systems is generally reported in terms of the dual measures of precision and recalP It is fairly standard to report precision and recall scores in the mid to high 90% range.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The method shares some characteristics of the decision list algorithm presented in this paper.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The entire treebank is used in the experiment, but only primary dependencies are considered.4 In all experiments, punctuation tokens are included in the data but omitted in evaluation scores.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"For the experiments, we use a simple preprocessing step.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Figure 2 shows a screenshot (which is of somewhat limited value, though, as color plays a major role in signalling the different statuses of the information).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The resulting algorithm is depicted in Table 1.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We indicate whether a context with zero log backoff will extend using the sign bit: +0.0 for contexts that extend and \u22120.0 for contexts that do not extend.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We assume that M is in an existential state qA, with integers i1 and i2 representing zi in the (2i \u2014 1)th and 22th work tape, for 1 < i < k. For each rule p : A fp(B, C) such that fp is mapped onto the function fp defined by the following rule. jp((xi,.. \u2022 ,rnt), (1ii, \u2022 \u2022 \u2022 \u2022 Yn3))= (Zi , \u2022 \u2022 \u2022 , Zk) M breaks xi , zk into substrings xi, , xn, and yi,...,y&quot; conforming to the definition of fp.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"10 Chinese speakers may object to this form, since the suffix f, menD (PL) is usually restricted to.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Our results suggest that current parsing models would benefit from better annotation consistency and enriched annotation in certain syntactic configurations.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This decreases the statistical significance of our results compared to those studies.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The segmenter will give both analyses :1 cai2 neng2 'just be able,' and ?]cai2neng2 'talent,' but the latter analysis is preferred since splitting these two morphemes is generally more costly than grouping them.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We propose maximizing , where is the sequence of named- entity tags assigned to the words in the sentence , and is the information that can be extracted from the whole document containing . Our system is built on a maximum entropy classifier.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Each word is terminated by an arc that represents the transduction between f and the part of speech of that word, weighted with an estimated cost for that word.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For each pair of judges consider one judge as the standard,.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"1 55.8 38.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Since the early days of statistical NLP, researchers have observed that a part-of-speech tag distribution exhibits \u00e2\u0080\u009cone tag per discourse\u00e2\u0080\u009d sparsity \u00e2\u0080\u0094 words are likely to select a single predominant tag in a corpus, even when several tags are possible.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This is in general very difficult, given the extremely free manner in which Chinese given names are formed, and given that in these cases we lack even a family name to give the model confidence that it is identifying a name.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"These knowledge sources determine whether the contexts surrounding an anaphor and antecedent are compatible.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Our system fails in (a) because of$ shenl, a rare family name; the system identifies it as a family name, whereas it should be analyzed as part of the given name.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"97 81.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Much could be done to further reduce memory consumption.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In this paper we present a stochastic finite-state model for segmenting Chinese text into words, both words found in a (static) lexicon as well as words derived via the above-mentioned productive processes.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"However, in order to capture the properties of various grammatical systems under consideration, our notation is more restrictive that ILFP, which was designed as a general logical notation to characterize the complete class of languages that are recognizable in polynomial time.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"), and thosethat begin with a verb (\u00ef\u00bf\u00bd ub..i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In IdentiFinder, there is a priority in the feature assignment, such that if one feature is used for a token, another feature lower in priority will not be used.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Particular instances of relations are associated with goodness scores.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"(Carlson, Marcu 2001) responded to this situation with relatively precise (and therefore long!)","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"At this step, we will try to link those sets, and put them into a single cluster.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":".., Wn ) (obs) P T : Tag assigns (T1 ,.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Trees are composed using an operation called adjoining, which is defined as follows.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This is to allow for fair comparison between the statistical method and GR, which is also purely dictionary-based.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Contains digits.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The rest of the paper is structured as follows.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"and f,.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"ATB CTB6 Negra WSJ Trees 23449 28278 20602 43948 Word Typess 40972 45245 51272 46348 Tokens 738654 782541 355096 1046829 Tags 32 34 499 45 Phrasal Cats 22 26 325 27 Test OOV 16.8% 22.2% 30.5% 13.2% Per Sentence Table 4: Gross statistics for several different treebanks.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thus, the length of any string in L is a linear combination of the length of strings in some fixed finite subset of L, and thus L is said to have the constant growth property.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The CoBoost algorithm described above divides the function Zco into two parts: Zco = 40 + 40.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"There is no global pruning.","target_text":"2.1 Inverted Alignments.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This approach makes the training objective more complex by adding linear constraints proportional to the number of word types, which is rather prohibitive.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"A different but supplementary perspective on discourse-based information structure is taken 11ventionalized patterns (e.g., order of informa by one of our partner projects, which is inter tion in news reports).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"3.1 Gross Statistics.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Here, the pruning threshold t0 = 10:0 is used.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Subsets C of increasing cardinality c are processed.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Though we are not able to calculate their memory usage on our model, results reported in their paper suggest lower memory consumption than TRIE on large-scale models, at the expense of CPU time.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Before we turn to the evaluation, however, we need to introduce the data-driven dependency parser used in the latter experiments.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"11 www.ling.unipotsdam.de\/sfb\/projekt a3.php 12 This step was carried out in the course of the diploma thesis work of David Reitter (2003), which de serves special mention here.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"On a set of 11 sentence fragments-the A set-where they reported 100% recall and precision for name identification, we had 73% recall and 80% precision.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"As mentioned above, it is not obvious how to apply Daum\u00b4e\u2019s approach to multinomials, which do not have a mechanism for combining split features.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For all lists except locations, the lists are processed into a list of tokens (unigrams).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The accuracy is calculated as the ratio of the number of paraphrases to the total number of phrases in the set.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We compared the ATB5 to tree- banks for Chinese (CTB6), German (Negra), and English (WSJ) (Table 4).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Out of those 15 links, 4 are errors, namely \u00e2\u0080\u009cbuy - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cpurchase - stake\u00e2\u0080\u009d \u00e2\u0080\u009cacquisition - stake\u00e2\u0080\u009d.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"ni2ya3 and @5:2 xilya3, respectively.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In the namedentity problem each example is a (spelling,context) pair.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"\u00e2\u0080\u009d, because Bush is the first word, the initial caps might be due to its position (as in \u00e2\u0080\u009cThey put a freeze on . . .","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For example, in Northern dialects (such as Beijing), a full tone (1, 2, 3, or 4) is changed to a neutral tone (0) in the final syllable of many words: Jll donglgual 'winter melon' is often pronounced donglguaO.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"REL+VB) (cf.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The first term in the objective function is the graph smoothness regularizer which encourages the distributions of similar vertices (large wij) to be similar.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The first is an evaluation of the system's ability to mimic humans at the task of segmenting text into word-sized units; the second evaluates the proper-name identification; the third measures the performance on morphological analysis.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We have not explored this strategy.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"2 We now introduce a new algorithm for learning from unlabeled examples, which we will call DLCoTrain (DL stands for decision list, the term Cotrain is taken from (Blum and Mitchell 98)).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We would also like to thank Amarnag Subramanya for helping us with the implementation of label propagation and Shankar Kumar for access to the parallel data.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Entries for 2 < n < N store a vocabulary identifier, probability, backoff, and an index into the n + 1-gram table.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The approach uses both spelling and contextual rules.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"While size of the resulting transducers may seem daunting-the segmenter described here, as it is used in the Bell Labs Mandarin TTS system has about 32,000 states and 209,000 arcs-recent work on minimization of weighted machines and transducers (cf.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Hence, trees shown in Figure 8 can not be generated by any MCTAG (but can be generated by an IG) because the number of pairs of dependent paths grows with n. Since the derivation tees of TAG's, MCTAG's, and HG's are local sets, the choice of the structure used at each point in a derivation in these systems does not depend on the context at that point within the derivation.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"2.1 Inverted Alignments.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"To improve agreement during the revision process, a dual-blind evaluation was performed in which 10% of the data was annotated by independent teams.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(4) is minimized by setting Since a feature may be present in only a few examples, W_ can be in practice very small or even 0, leading to extreme confidence values.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Input: Ja , wunderbar . Konnen wir machen . MonS: Yes, wonderful.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For pronouns, however, all of the knowledge sources increased recall, often substantially, and with little if any decrease in precision.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Time for Moses itself to load, including loading the language model and phrase table, is included.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The government has to make a decision, and do it quickly.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In order to solve this problem, a parse tree is needed to understand that \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is not the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In the following, we assume that this word joining has been carried out.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Statistical methods seem particularly applicable to the problem of unknown-word identification, especially for constructions like names, where the linguistic constraints are minimal, and where one therefore wants to know not only that a particular se\u00c2\u00ad quence of hanzi might be a name, but that it is likely to be a name with some probabil\u00c2\u00ad ity.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Our full model outperforms the \u201cNo LP\u201d setting because it has better vocabulary coverage and allows the extraction of a larger set of constraint features.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"These performance gains transfer to improved system runtime performance; though we focused on Moses, our code is the best lossless option with cdec and Joshua.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"As is described in Section 4, the property of having a derivation tree set that is a local set appears to be useful in showing important properties of the languages generated by the formalisms.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"At the phrasal level, we remove all function tags and traces.","label":0},{"source_text":"There is no global pruning.","target_text":"The sequence of states needed to carry out the word reordering example in Fig.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"German city names are replaced by category markers.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"attaching to terms denoting human beings.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Otherwise, it is set to 0.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"2.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"summit Sharm (a) Al-Sheikh summit Sharm (b) DTNNP Al-Sheikh in a corpus position without a bracketing label, then we also add \u00e2\u0088\u0097n, NIL) to M. We call the set of unique n-grams with multiple labels in M the variation nuclei of C. Bracketing variation can result from either annotation errors or linguistic ambiguity.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"There are thus some very good reasons why segmentation into words is an important task.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Figure 1 shows sample sentences from these domains, which are widely divergent.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In contrast to these approaches, our method directly incorporates these constraints into the structure of the model.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Personal names such as 00, 3R; zhoulenl-lai2 'Zhou Enlai.'","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Only 2 link in the CC- domain (buy-purchase, acquire-acquisition) and 2 links (trader-dealer and head-chief) in the PC- domain are found in the same synset of Word- Net 2.1 (http:\/\/wordnet.princeton.edu\/).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For example, a good classifier would identify Mrs. Frank as a person, Steptoe & Johnson as a company, and Honduras as a location.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"8 1 8.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"In general, the neighborhoods can be more diverse and we allow a soft label distribution over the vertices.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"As described in Sproat (1995), the Chinese segmenter presented here fits directly into the context of a broader finite-state model of text analysis for speech synthesis.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This is a rather important source of errors in name identifi\u00c2\u00ad cation, and it is not really possible to objectively evaluate a name recognition system without considering the main lexicon with which it is used.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Although the segmentation requirements for Arabic are not as extreme as those for Chinese, Arabic is written with certain cliticized prepositions, pronouns, and connectives connected to adjacent words.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Since the transducers are built from human-readable descriptions using a lexical toolkit (Sproat 1995), the system is easily maintained and extended.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"For each source word f, the list of its possible translations e is sorted according to p(fje) puni(e), where puni(e) is the unigram probability of the English word e. It is su\u00c3\u0086cient to consider only the best 50 words.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Of these cases, 38 were temporal expressions (either a day of the week or month of the year).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The morphological anal\u00c2\u00adysis itself can be handled using well-known techniques from finite-state morphol 9 The initial estimates are derived from the frequencies in the corpus of the strings of hanzi making up.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In particular, it may not be possible to learn functions fi (x f2(x2,t) for i = m + 1...n: either because there is some noise in the data, or because it is just not realistic to expect to learn perfect classifiers given the features used for representation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, the characterization given in the main body of the text is correct sufficiently often to be useful.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Tag set As is standard, for all experiments, we set the number of latent model tag states to the size of the annotated tag set.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"In each figure the upper graph shows the isolated constituent precision and the bottom graph shows the corresponding number of hypothesized constituents.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"REL+VB) (cf.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"67 95.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"With the same Corporate- Suffix-List and Person-Prefix-List used in local features, for a token seen elsewhere in the same document with one of these suffixes (or prefixes), another feature Other-CS (or Other-PP) is set to 1.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":1},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Other good classes include JADE and GOLD; other bad classes are DEATH and RAT.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Unfortunately, modifying the model to account for these kind of dependencies is not at all straightforward.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Thus, the arc from je to jedna will be labeled 5b\u2193 (to indicate that there is a syntactic head below it).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Its only purpose is 3 This follows since each \u00ce\u00b8t has St \u00e2\u0088\u0092 1 parameters and.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The reference medicine for Silapo is EPREX\/ERYPO, which contains epoetin alfa.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The problem can be represented as a graph with 2N vertices corresponding to the members of X1 and X2.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"In an experiment on automatic rhetorical parsing, the RST-annotations and PoS tags were used by (Reitter 2003) as a training corpus for statistical classification with Support Vector Machines.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"For the remaining arcs, if the segment is in fact a known lexeme it is tagged as usual, but for the OOV arcs which are valid Hebrew entries lacking tags assignment, we assign all possible tags and then simulate a grammatical constraint.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"This number must be less than or equal to n \u00f4\u0080\u0080\u0080 1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Foreign names are usually transliterated using hanzi whose sequential pronunciation mimics the source language pronunciation of the name.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"All notions of word, with the exception of the orthographic word, are as relevant in Chinese as they are in English, and just as is the case in other languages, a word in Chinese may correspond to one or more symbols in the orthog 1 For a related approach to the problem of word-segrnention in Japanese, see Nagata (1994), inter alia..","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Figure 1 reveals that an event that \u00e2\u0080\u009cdamaged\u00e2\u0080\u009d objects may also cause injuries; a disaster that \u00e2\u0080\u009coccurred\u00e2\u0080\u009d may be investigated to find its \u00e2\u0080\u009ccause\u00e2\u0080\u009d; a disaster may \u00e2\u0080\u009cwreak\u00e2\u0080\u009d havoc as it \u00e2\u0080\u009ccrosses\u00e2\u0080\u009d geographic regions; and vehicles that have a \u00e2\u0080\u009cdriver\u00e2\u0080\u009d may also \u00e2\u0080\u009ccarry\u00e2\u0080\u009d items.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"For all languages, the vocabulary sizes increase by several thousand words.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"This might be because our features are more comprehensive than those used by Borthwick.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We first give a brief overview of boosting algorithms.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The first modification \u2014 cautiousness \u2014 is a relatively minor change.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Table 1 The cost as a novel given name (second position) for hanzi from various radical classes.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The computation of Pfr1(c)1Mi M k (C)) has been sketched before in Equations 1 through 4.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Because of this threshold, very few NE instance pairs could be used and hence the variety of phrases was also limited.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"An initial step of any text\u00c2\u00ad analysis task is the tokenization of the input into words.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"KS Function Ge nde r filters candidate if gender doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Particular instances of relations are associated with goodness scores.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In this paper, we make a simplifying assumption of one-tag-per-word.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each co-occurrence relation (noun\/caseframe for CFLex, and caseframe\/caseframe for CFNet), BABAR computes its log-likelihood value and looks it up in the \u00cf\u00872 table to obtain a confidence level.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Label propagation is used to propagate these tags inwards and results in tag distributions for the middle word of each Italian trigram.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Examples will usually be accompanied by a translation, plus a morpheme-by-morpheme gloss given in parentheses whenever the translation does not adequately serve this purpose.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"to explore how well we can induce POS tags using only the one-tag-per-word constraint.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Finally, note that while most feature concepts are lexicalized, others, such as the suffix concept, are not.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Pseudo-labels are formed by taking seed labels on the labeled examples, and the output of the fixed classifier on the unlabeled examples.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Although the best published results for the Collins parser is 80% UAS (Collins, 1999), this parser reaches 82% when trained on the entire training data set, and an adapted version of Charniak\u2019s parser (Charniak, 2000) performs at 84% (Jan Haji\u02c7c, pers. comm.).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"For instance, by altering the emission distribution parameters, Johnson (2007) encourages the model to put most of the probability mass on few tags.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"42 nator, the N31s can be measured well by counting, and we replace the expectation by the observation.","label":0},{"source_text":"There is no global pruning.","target_text":"Although the ultimate goal of the Verbmobil project is the translation of spoken language, the input used for the translation experiments reported on in this paper is the (more or less) correct orthographic transcription of the spoken sentences.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"We show translation results for three approaches: the monotone search (MonS), where no word reordering is allowed (Tillmann, 1997), the quasimonotone search (QmS) as presented in this paper and the IBM style (IbmS) search as described in Section 3.2.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For novel texts, no lexicon that consists simply of a list of word entries will ever be entirely satisfactory, since the list will inevitably omit many constructions that should be considered words.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This is most severe with RandLM in the multi-threaded case, where each thread keeps a separate cache, exceeding the original model size.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The BerkeleyLM direct-mapped cache is in principle faster than caches implemented by RandLM and by IRSTLM, so we may write a C++ equivalent implementation as future work.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"This can be repeated several times to collect a list of author \/ book title pairs and expressions.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"ments contained 322 anaphoric links.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Several papers report the use of part-of-speech information to rank segmentations (Lin, Chiang, and Su 1993; Peng and Chang 1993; Chang and Chen 1993); typically, the probability of a segmentation is multiplied by the probability of the tagging(s) for that segmentation to yield an estimate of the total probability for the analysis.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We also thank ChaoHuang Chang, reviewers for the 1994 ACL conference, and four anonymous reviewers for Computational Linguistics for useful comments.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Table 3: Training and test conditions for the Verbmobil task (*number of words without punctuation marks).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"pre-processing.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Moses keeps language models and many other resources in static variables, so these are still resident in memory.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"These 140 NE categories are designed by extending MUC\u00e2\u0080\u0099s 7 NE categories with finer sub-categories (such as Company, Institute, and Political Party for Organization; and Country, Province, and City for Location) and adding some new types of NE categories (Position Title, Product, Event, and Natural Object).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For sorted lookup, we compare interpolation search, standard C++ binary search, and standard C++ set based on red-black trees.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Previous Work.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"A different knowledge source, called CFSemCFSem, compares the semantic expectations of the caseframe that extracts the anaphor with the semantic expectations of the caseframe that extracts the candidate.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We present a coreference resolver called BABAR that uses contextual role knowledge to evaluate possible antecedents for an anaphor.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"A totally non\u00c2\u00ad stochastic rule-based system such as Wang, Li, and Chang's will generally succeed in such cases, but of course runs the risk of overgeneration wherever the single-hanzi word is really intended.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"SRILM\u2019s compact variant, IRSTLM, MITLM, and BerkeleyLM\u2019s sorted variant are all based on this technique.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"3 These are not full case frames in the traditional sense, but they approximate a simple case frame with a single slot.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"1.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"For each token, three types of features may be taken into account: the word form; the part-of-speech assigned by an automatic tagger; and labels on previously assigned dependency arcs involving the token \u2013 the arc from its head and the arcs to its leftmost and rightmost dependent, respectively.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Top 5 Bot to m 5 Go ld NN P NN JJ CD NN S RB S PD T # \u00e2\u0080\u009d , 1T W CD W RB NN S VB N NN PR P$ W DT : MD . +P RI OR CD JJ NN S WP $ NN RR B- , $ \u00e2\u0080\u009d . +F EA TS JJ NN S CD NN P UH , PR P$ # . \u00e2\u0080\u009c Table 5: Type-level English POS Tag Ranking: We list the top 5 and bottom 5 POS tags in the lexicon and the predictions of our models under the best hyperparameter setting.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As described in Sproat (1995), the Chinese segmenter presented here fits directly into the context of a broader finite-state model of text analysis for speech synthesis.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Our use of an unweighted lattice reflects our belief that all the segmentations of the given input sentence are a-priori equally likely; the only reason to prefer one segmentation over the another is due to the overall syntactic context which is modeled via the PCFG derivations.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We evaluate our approach on seven languages: English, Danish, Dutch, German, Portuguese, Spanish, and Swedish.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This enables us to build a high performance NER without using separate classifiers to take care of global consistency or complex formulation on smoothing and backoff models (Bikel et al., 1997).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If all three sentences are in the same document, then even a human will find it difficult to classify McCann in (1) into either person or organization, unless there is some other information provided.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Under this scheme, n human judges are asked independently to segment a text.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"All the sentences have been analyzed by our chunker and NE tag- ger.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Suffixes and Prefixes: This group contains only two features: Corporate-Suffix and Person-Prefix.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"2.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The edge weights between the foreign language trigrams are computed using a co-occurence based similarity function, designed to indicate how syntactically similar the middle words of the connected trigrams are (\u00a73.2).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Moreover, they are used as substantives much 2 Unlike machine translation, constituency parsing is not significantly affected by variable word order.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Following Dickinson (2005), we randomly sampled 100 variation nuclei from each corpus and evaluated each sample for the presence of an annotation error.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"(a) Of the high frequency phrasal categories, ADJP and SBAR are the hardest to parse.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In particular, it may not be possible to learn functions fi (x f2(x2,t) for i = m + 1...n: either because there is some noise in the data, or because it is just not realistic to expect to learn perfect classifiers given the features used for representation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This suggests that the backoff model is as reasonable a model as we can use in the absence of further information about the expected cost of a plural form.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"3.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"As is standard, we report the greedy one-to-one (Haghighi and Klein, 2006) and the many-to-one token-level accuracy obtained from mapping model states to gold POS tags.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Restrictions We compare our new approach with the word reordering used in the IBM translation approach (Berger et al., 1996).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The problem of &quot;noise&quot; items that do not fall into any of the three categories also needs to be addressed.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Extract NE instance pairs with contexts First, we extract NE pair instances with their context from the corpus.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We showed in \u00c2\u00a72 that lexical ambiguity explains the underperformance of these categories.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Figure 2 shows examples of lexical expectations that were learned for both domains.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We received submissions from 14 groups from 11 institutions, as listed in Figure 2.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Decreasing the threshold results in higher mWER due to additional search errors.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The corpora for both settings are summarized in table 1.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We have studied the structural descriptions (tree sets) that can be assigned by various grammatical systems, and classified these formalisms on the basis of two features: path complexity; and path independence.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We first give a brief overview of boosting algorithms.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"It can be seen that M performs a top-down recognition of the input al ... nin logspace.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In (Bikel et al., 1997) and (Bikel et al., 1999), performance was plotted against training data size to show how performance improves with training data size.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The idea is to have a pipeline of shallow-analysis modules (tagging, chunk- ing, discourse parsing based on connectives) and map the resulting underspecified rhetorical tree (see Section 2.4) into a knowledge base that may contain domain and world knowledge for enriching the representation, e.g., to resolve references that cannot be handled by shallow methods, or to hypothesize coherence relations.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Thus corresponding pseudo-labels for instances on which gj abstain are set to zero and these instances do not contribute to the objective function.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"If the context wnf will never extend to the right (i.e. wnf v is not present in the model for all words v) then no subsequent query will match the full context.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The paper is structured as follows.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Future extensions of the system might include: 1) An extended translation model, where we use more context to predict a source word.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"After adding a ROOT node to all trees, we train a grammar using six split-and- merge cycles and no Markovization.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Using the concept of inverted alignments, we explicitly take care of the coverage constraint by introducing a coverage set C of source sentence positions that have been already processed.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"In our case multi-threading is trivial because our data structures are read-only and uncached.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Second, BABAR performs reliable case resolution to identify anaphora that can be easily resolved using the lexical and syntactic heuristics described in Section 2.1.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Following the convention presented in earlier sections, we assume that each example is an instance pair of the from (xi ,i, x2,) where xj,, E 2x3 , j E 2}.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"By aggregating the POS labels of the English tokens to types, we can generate label distributions for the English vertices.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Table 1 shows results of the benchmark.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"When the parser is trained on the transformed data, it will ideally learn not only to construct projective dependency structures but also to assign arc labels that encode information about lifts.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The following features were used: full-string=x The full string (e.g., for Maury Cooper, full- s tring=Maury_Cooper). contains(x) If the spelling contains more than one word, this feature applies for any words that the string contains (e.g., Maury Cooper contributes two such features, contains (Maury) and contains (Cooper) . allcapl This feature appears if the spelling is a single word which is all capitals (e.g., IBM would contribute this feature). allcap2 This feature appears if the spelling is a single word which is all capitals or full periods, and contains at least one period.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Finally, quite a few hanzi are homographs, meaning that they may be pronounced in several different ways, and in extreme cases apparently represent different morphemes: The prenominal modifi\u00c2\u00ad cation marker eg deO is presumably a different morpheme from the second morpheme of \u00c2\u00a7eg mu4di4, even though they are written the same way.4 The second point, which will be relevant in the discussion of personal names in Section 4.4, relates to the internal structure of hanzi.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But it conflates the coordinating and discourse separator functions of wa (<..4.b \u00ef\u00bf\u00bd \u00ef\u00bf\u00bd) into one analysis: conjunction(Table 3).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"information structure.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"None of the models attach the attributive adjectives correctly.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In our experiments we set the parameter values randomly, and then ran EM to convergence.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"If these sets do not overlap, then the words cannot be coreferent.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"All of these systems were run on data that was not seen during their development.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For example, given a sequence F1G1G2, where F1 is a legal single-hanzi family name, and Plural Nouns X g 0 g \"' X X 0 T!i c\"'.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"As expected, the most informative encoding, Head+Path, gives the highest accuracy with over 99% of all non-projective arcs being recovered correctly in both data sets.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The correct resolution in sentence (b) comes from knowledge that people who are kidnapped are often subsequently released.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We compared the ATB5 to tree- banks for Chinese (CTB6), German (Negra), and English (WSJ) (Table 4).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For our demonstration system, we typically use the pruning threshold t0 = 5:0 to speed up the search by a factor 5 while allowing for a small degradation in translation accuracy.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Table 5 provides insight into the behavior of different models in terms of the tagging lexicon they generate.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Upon identifying an anaphoric expression (currently restricted to: pronouns, prepositional adverbs, definite noun phrases), the an- notator first marks the antecedent expression (currently restricted to: various kinds of noun phrases, prepositional phrases, verb phrases, sentences) and then establishes the link between the two.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The logistic function, whose outputs are in [0, 1], forces pp(s, t) <_ po(s, t).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For example Chen and Liu (1992) report precision and recall rates of over 99%, but this counts only the words that occur in the test corpus that also occur in their dictionary.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Additionally, for a given coverage set, at most 250 different hypotheses are kept during the search process, and the number of different words to be hypothesized by a source word is limited.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Future work should also extend the approach to build a complete named entity extractor - a method that pulls proper names from text and then classifies them.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Figure 4 Input lattice (top) and two segmentations (bottom) of the sentence 'How do you say octopus in Japanese?'.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In this section, we describe the seven general knowledge sources and explain how the DempsterShafer model makes resolutions.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"For developers of Statistical Machine Translation (SMT) systems, an additional complication is the heterogeneous nature of SMT components (word-alignment model, language model, translation model, etc.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"There are two key benefits of this model architecture.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Each learner is free to pick the labels for these instances.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Finally, the statistical method fails to correctly group hanzi in cases where the individual hanzi comprising the name are listed in the dictionary as being relatively high-frequency single-hanzi words.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"With the same Corporate- Suffix-List and Person-Prefix-List used in local features, for a token seen elsewhere in the same document with one of these suffixes (or prefixes), another feature Other-CS (or Other-PP) is set to 1.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The paper is organized as follows: Section 2 explains the different layers of annotation that have been produced or are being produced.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The features we used can be divided into 2 classes: local and global.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Language models are widely applied in natural language processing, and applications such as machine translation make very frequent queries.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The problem with these styles of evaluation is that, as we shall demonstrate, even human judges do not agree perfectly on how to segment a given text.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","label":1},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Because the information flow in our graph is asymmetric (from English to the foreign language), we use different types of vertices for each language.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The alignment model uses two kinds of parameters: alignment probabilities p(aj jaj\u00f4\u0080\u0080\u00801; I; J), where the probability of alignment aj for position j depends on the previous alignment position aj\u00f4\u0080\u0080\u00801 (Ney et al., 2000) and lexicon probabilities p(fj jeaj ).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Our baseline for all sentence lengths is 5.23% F1 higher than the best previous result.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"However, we note that the final conditional estimates p(s|t) from a given phrase table maximize the likelihood of joint empirical phrase pair counts over a word-aligned corpus.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For fair comparison, we have tabulated all results with the size of training data used (Table 5 and Table 6).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"To be short, we omit the target words e; e0 in the formulation of the search hypotheses.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If a candidate has a belief value \u00e2\u0089\u00a5 .50, then we select that candidate as the antecedent for the anaphor.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The final estimating equation is then: (3) Since the total of all these class estimates was about 10% off from the Turing estimate n1\/N for the probability of all unseen hanzi, we renormalized the estimates so that they would sum to n 1jN.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"In our third model GTppp we also add the distinction between general PPs and possessive PPs following Goldberg and Elhadad (2007).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"They are also labelled for their topicality (yes \/ no), and this annotation is accompanied by a confidence value assigned by the annotator (since it is a more subjective matter).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Figure 3 shows a small fragment of the WFST encoding the dictionary, containing both entries forjust discussed, g:t\u00c2\u00a5 zhonglhua2 min2guo2 (China Republic) 'Republic of China,' and i\u00c2\u00a5inl.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The rhetorical structure annotations of PCC have all been converted to URML.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"A defining characteristic of MSA is the prevalence of discourse markers to connect and subordinate words and phrases (Ryding, 2005).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Segmentation errors cascade into the parsing phase, placing an artificial limit on parsing performance.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The morphological anal\u00c2\u00adysis itself can be handled using well-known techniques from finite-state morphol 9 The initial estimates are derived from the frequencies in the corpus of the strings of hanzi making up.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Proper names that match are resolved with each other.","label":0},{"source_text":"Here we present two algorithms.","target_text":"2 for the accuracy of the different methods.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In our forth model GTnph we add the definiteness status of constituents following Tsarfaty and Sima\u2019an (2007).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"For a corpus C, let M be the set of tuples \u00e2\u0088\u0097n, l), where n is an n-gram with bracketing label l. If any n appears 6 Generative parsing performance is known to deteriorate with sentence length.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"To show that the derivation trees of any grammar in LCFRS is a local set, we can rewrite the annotated derivation trees such that every node is labelled by a pair to include the composition operations.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"4 7 . 3 8 . 9 2 8 . 8 2 0 . 7 3 2 . 3 3 5 . 2 2 9 . 6 2 7 . 6 1 4 . 2 4 2 . 8 4 5 . 9 4 4 . 3 6 0 . 6 6 1 . 5 4 9 . 9 3 3 . 9 Table 6: Type-level Results: Each cell report the type- level accuracy computed against the most frequent tag of each word type.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Each member of a set of trees can be adjoined into distinct nodes of trees in a single elementary tree set, i.e, derivations always involve the adjunction of a derived auxiliary tree set into an elementary tree set.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In the case of, the most common usage is as an adverb with the pronunciation jiangl, so that variant is assigned the estimated cost of 5.98, and a high cost is assigned to nominal usage with the pronunciation jiang4.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"As a result, Habash et al.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The unlabeled data gives many such &quot;hints&quot; that two features should predict the same label, and these hints turn out to be surprisingly useful when building a classifier.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In the present study, we limit ourselves to an algorithmic approach, using a deterministic breadthfirst search.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"32 81.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Note that since all English vertices were extracted from the parallel text, we will have an initial label distribution for all vertices in Ve.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We obtained positive results using a very simple phrase-based system in two different adaptation settings: using English\/French Europarl to improve a performance on a small, specialized medical domain; and using non-news portions of the NIST09 training material to improve performance on the news-related corpora.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We map the ATB morphological analyses to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags for all experiments.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"5 67.3 55.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"On average, 6 reference translations per automatic translation are available.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"na me =>2 ha nzi fa mi ly 2 ha nzi gi ve n 5.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Each word is terminated by an arc that represents the transduction between f and the part of speech of that word, weighted with an estimated cost for that word.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In this paper, we make a simplifying assumption of one-tag-per-word.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"These derivation trees show how the composition operations were used to derive the final structures from elementary structures.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Evaluation of the Segmentation as a Whole.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"In practice, this sparsity constraint is difficult to incorporate in a traditional POS induction system (Me\u00c2\u00b4rialdo, 1994; Johnson, 2007; Gao and Johnson, 2008; Grac\u00c2\u00b8a et al., 2009; Berg-Kirkpatrick et al., 2010).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"A more forceful approach for encoding sparsity is posterior regularization, which constrains the posterior to have a small number of expected tag assignments (Grac\u00c2\u00b8a et al., 2009).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The corpus-based statistical parsing community has many fast and accurate automated parsing systems, including systems produced by Collins (1997), Charniak (1997) and Ratnaparkhi (1997).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Each out-of-domain phrase pair is characterized by a set of simple features intended to reflect how useful it will be.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"These alignment models are similar to the concept of hidden Markov models (HMM) in speech recognition.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"For example, the phrase \u00e2\u0080\u009c's New York-based trust unit,\u00e2\u0080\u009d is not a paraphrase of the other phrases in the \u00e2\u0080\u009cunit\u00e2\u0080\u009d set.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Each out-of-domain phrase pair is characterized by a set of simple features intended to reflect how useful it will be.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Following the setup of Johnson (2007), we use the whole of the Penn Treebank corpus for training and evaluation on English.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Morphological processes in Semitic languages deliver space-delimited words which introduce multiple, distinct, syntactic units into the structure of the input sentence.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The method halves the error rate in comparison to a method using the labeled examples alone.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For each pair of judges, consider one judge as the standard,.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We reduce this to O(log log |A|) time by evenly distributing keys over their range then using interpolation search4 (Perl et al., 1978).","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In the labeled version of these metrics (L) both heads and arc labels must be correct, while the unlabeled version (U) only considers heads.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Both implementations employ a state object, opaque to the application, that carries information from one query to the next; we discuss both further in Section 4.2.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"BABAR then computes statistics over the training examples measuring the frequency with which extraction patterns and noun phrases co-occur in coreference resolutions.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"3.2 Results.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The parallel data came from the Europarl corpus (Koehn, 2005) and the ODS United Nations dataset (UN, 2006).","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"For developing these mechanisms, the possibility to feed in hand-annotated information is very useful.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Its only purpose is 3 This follows since each \u00ce\u00b8t has St \u00e2\u0088\u0092 1 parameters and.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Additionally, for a given coverage set, at most 250 different hypotheses are kept during the search process, and the number of different words to be hypothesized by a source word is limited.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Figure 3 shows examples of semantic expectations that were learned.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"A typical situation is shown in Figure 1.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The particular classifier used depends upon the noun.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Even if an example like this is not labeled, it can be interpreted as a &quot;hint&quot; that Mr and president imply the same category.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Here we use a slightly different notion of lift, applying to individual arcs and moving their head upwards one step at a time: Intuitively, lifting an arc makes the word wk dependent on the head wi of its original head wj (which is unique in a well-formed dependency graph), unless wj is a root in which case the operation is undefined (but then wj \u2014* wk is necessarily projective if the dependency graph is well-formed).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"There, a lattice is used to represent the possible sentences resulting from an interpretation of an acoustic model.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"7 Acknowledgements.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"It has been shown for English (Wang and Hirschberg 1992; Hirschberg 1993; Sproat 1994, inter alia) that grammatical part of speech provides useful information for these tasks.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We define a symmetric similarity function K(uZ7 uj) over two foreign language vertices uZ7 uj E Vf based on the co-occurrence statistics of the nine feature concepts given in Table 1.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For E(ni1s), then, we substitute a smooth S against the number of class elements.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"For the translation model Pr(fJ 1 jeI 1), we go on the assumption that each source word is aligned to exactly one target word.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The monotone search performs worst in terms of both error rates mWER and SSER.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"We often encounter sentences that are highly ambiguous in themselves, without some prior knowledge of the entities concerned.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The work of the first author was supported by the Lynn and William Frankel Center for Computer Sciences.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Most computational models for coreference resolution rely on properties of the anaphor and candidate antecedent, such as lexical matching, grammatical and syntactic features, semantic agreement, and positional information.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The independence of paths in the tree sets of the k tI grammatical formalism in this hierarchy can be shown by means of tree pumping lemma of the form t1ti3t .","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"input: source string f1:::fj :::fJ initialization for each cardinality c = 1; 2; ; J do for each pair (C; j), where j 2 C and jCj = c do for each target word e 2 E Qe0 (e; C; j) = p(fj je) max \u00c3\u0086;e00 j02Cnfjg fp(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j0)g words fj in the input string of length J. For the final translation each source position is considered exactly once.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, if CFLex determines that the log- likelihood statistic for the co-occurrence of a particular noun and caseframe corresponds to the 90% confidence level, then CFLex returns .90 as its belief that the anaphor and candidate are coreferent.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"This distributional sparsity of syntactic tags is not unique to English 1 The source code for the work presented in this paper is available at http:\/\/groups.csail.mit.edu\/rbg\/code\/typetagging\/.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Put another way, the minimum of Equ.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"computing the recall of the other's judgments relative to this standard.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The algorithm can be viewed as heuristically optimizing an objective function suggested by (Blum and Mitchell 98); empirically it is shown to be quite successful in optimizing this criterion.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Much could be done to further reduce memory consumption.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The learning task is to find two classifiers : 2x1 { \u20141, +1} 12 : 2x2 { \u20141, +1} such that (x1,) = f2(x2,t) = yt for examples i = 1, , m, and f1 (x1,) = f2 (x2,t) as often as possible on examples i = m + 1, ,n. To achieve this goal we extend the auxiliary function that bounds the training error (see Equ.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The AdaBoost algorithm was developed for supervised learning.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Current state-of-the-art broad-coverage parsers assume a direct correspondence between the lexical items ingrained in the proposed syntactic analyses (the yields of syntactic parse-trees) and the spacedelimited tokens (henceforth, \u2018tokens\u2019) that constitute the unanalyzed surface forms (utterances).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Preferences for constituent order (especially in languages with relatively free word order) often belong to this group.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Finally, we provide a realistic eval uation in which segmentation is performed both in a pipeline and jointly with parsing (\u00c2\u00a76).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The cost is computed as follows, where N is the corpus size and f is the frequency: (1) Besides actual words from the base dictionary, the lexicon contains all hanzi in the Big 5 Chinese code\/ with their pronunciation(s), plus entries for other characters that can be found in Chinese text, such as Roman letters, numerals, and special symbols.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"1 53.8 47.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"During development, we sensed that the Recency and Syn- role KSs did not deserve to be on equal footing with the other KSs because their knowledge was so general.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The first term in the objective function is the graph smoothness regularizer which encourages the distributions of similar vertices (large wij) to be similar.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"To lower the barrier of entrance to the competition, we provided a complete baseline MT system, along with data resources.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Experiments are presented in section 4.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In this work we extended the AdaBoost.MH (Schapire and Singer 98) algorithm to the cotraining case.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"paper, and is missing 6 examples from the A set.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Text generation, or at least the two phases of text planning and sentence planning, is a process driven partly by well-motivated choices (e.g., use this lexeme X rather than that more colloquial near-synonym Y ) and partly by con tation like that of PCC can be exploited to look for correlations in particular between syntactic structure, choice of referring expressions, and sentence-internal information structure.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Figure 1: Translation of PCC sample commentary (STTS)2.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The property of semilinearity is concerned only with the occurrence of symbols in strings and not their order.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If either case is true, then CFLex reports that the anaphor and candidate might be coreferent.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Search CPU time mWER SSER Method [sec] [%] [%] MonS 0:9 42:0 30:5 QmS 10:6 34:4 23:8 IbmS 28:6 38:2 26:2 4.2 Performance Measures.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"32 81.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Time starts when Moses is launched and therefore includes model loading time.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Set the decision list to include all rules whose (smoothed) strength is above some threshold Pmin.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We suspect that the higher precision in the disasters domain may be due to its substantially larger training corpus.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"With a human evaluation we also showed that ATB inter-annotator agreement remains low relative to the WSJ corpus.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"3 68.4 49.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In contrast, even though context can be helpful for resolving definite NPs, context can be trumped by the semantics of the nouns themselves.","label":0},{"source_text":"There is no global pruning.","target_text":"The traveling salesman problem is an optimization problem which is defined as follows: given are a set of cities S = s1; ; sn and for each pair of cities si; sj the cost dij > 0 for traveling from city si to city sj . We are looking for the shortest tour visiting all cities exactly once while starting and ending in city s1.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The algorithm in Fig.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The points enumerated above are particularly related to ITS, but analogous arguments can easily be given for other applications; see for example Wu and Tseng's (1993) discussion of the role of segmentation in information retrieval.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(5) and ht into Equ.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Modifying the Berkeley parser for Arabic is straightforward.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","label":1},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"These packages are further described in Section 3.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Twentieth-century linguistic work on Chinese (Chao 1968; Li and Thompson 1981; Tang 1988,1989, inter alia) has revealed the incorrectness of this traditional view.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"A simple extension will be used to handle this problem.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The 3rd block contains the mixture baselines.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Given that part-of-speech labels are properties of words rather than morphemes, it follows that one cannot do part-of-speech assignment without having access to word-boundary information.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For this example, since the sequence Even News Broadcasting Corp. only appears once in the document, its longest substring that occurs in the same document is News Broadcasting Corp. In this case, News has an additional feature of I begin set to 1, Broadcasting has an additional feature of I continue set to 1, and Corp. has an additional feature of I end set to 1.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE is then trained on 80% of the training corpus, and tested on the remaining 20%.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In Chinese text, individual characters of the script, to which we shall refer by their traditional name of hanzi,Z are written one after another with no intervening spaces; a Chinese sentence is shown in Figure 1.3 Partly as a result of this, the notion \"word\" has never played a role in Chinese philological tradition, and the idea that Chinese lacks any\u00c2\u00ad thing analogous to words in European languages has been prevalent among Western sinologists; see DeFrancis (1984).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(a) ;IE shi4 'be' => ;IE;IE shi4bu2-shi4 (be-not-be) 'is it?'","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Morphologically derived words such as, xue2shengl+men0.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"This research was supported in part by the Defense Advanced Research Projects Agency as part of the Translingual Information Detection, Extraction and Summarization (TIDES) program, under Grant N66001001-18917 from the Space and Naval Warfare Systems Center, San Diego, and by the National Science Foundation under Grant IIS00325657.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The final estimating equation is then: (3) Since the total of all these class estimates was about 10% off from the Turing estimate n1\/N for the probability of all unseen hanzi, we renormalized the estimates so that they would sum to n 1jN.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We then discuss how we adapt and generalize a boosting algorithm, AdaBoost, to the problem of named entity classification.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"13.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"After adding a ROOT node to all trees, we train a grammar using six split-and- merge cycles and no Markovization.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Step 1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Finally, we wish to reiterate an important point.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Otherwise, it is set to 0.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In Section 4, we present the performance measures used and give translation results on the Verbmobil task.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Tables 4 and 5 show BABAR\u00e2\u0080\u0099s performance when just one contextual role knowledge source is used at a time.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Unlike Kahane et al. (1998), we do not regard a projectivized representation as the final target of the parsing process.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The 2nd block contains the IR system, which was tuned by selecting text in multiples of the size of the EMEA training corpus, according to dev set performance.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"To our knowledge, ours is the first analysis of this kind for Arabic parsing.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Our experiments consistently demonstrate that this model architecture yields substantial performance gains over more complex tagging counterparts.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The first method uses a similar algorithm to that of (Yarowsky 95), with modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Note that Wang, Li, and Chang's.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The links can solve the problem.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This is orthographically represented as 7C.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Fig.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Many hanzi have more than one pronunciation, where the correct.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"To initialize the graph we tag the English side of the parallel text using a supervised model.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"BABAR uses a named entity recognizer to identify proper names that refer to people and companies.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In principle a feature could be an arbitrary predicate of the (spelling, context) pair; for reasons that will become clear, features are limited to querying either the spelling or context alone.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The algorithm builds two classifiers in parallel from labeled and unlabeled data.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"f;g denotes the empty set, where no source sentence position is covered.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For example, suppose one is building a ITS system for Mandarin Chinese.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Data We use the Hebrew Treebank, (Sima\u2019an et al., 2001), provided by the knowledge center for processing Hebrew, in which sentences from the daily newspaper \u201cHa\u2019aretz\u201d are morphologically segmented and syntactically annotated.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Otherwise, label the training data with the combined spelling\/contextual decision list, then induce a final decision list from the labeled examples where all rules (regardless of strength) are added to the decision list.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A high-level relation is agent, which relates an animate nominal to a predicate.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Note that the sets of possible classifiers for a given noun can easily be encoded on that noun by grammatical features, which can be referred to by finite-state grammatical rules.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The baseline alignment model does not permit that a source word is aligned to two or more target words, e.g. for the translation direction from German toEnglish, the German compound noun 'Zahnarztter min' causes problems, because it must be translated by the two target words dentist's appointment.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"our full model yields 39.3% average error reduction across languages when compared to the basic configuration (1TW).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The second modification is more important, and is discussed in the next section.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In the next section we present an alternative approach that builds two classifiers while attempting to satisfy the above constraints as much as possible.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This would result in better rest cost estimation and better pruning.10 In general, tighter, but well factored, integration between the decoder and language model should produce a significant speed improvement.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For instance, for TTS it is necessary to know that a particular sequence of hanzi is of a particular category because that knowl\u00c2\u00ad edge could affect the pronunciation; consider, for example the issues surrounding the pronunciation of ganl I qian2 discussed in Section 1.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In contrast, our morphological probabilities are based on a unigram, lexeme-based model, and all other (local and non-local) contextual considerations are delegated to the PCFG.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"We call this pseudoprojective dependency parsing, since it is based on a notion of pseudo-projectivity (Kahane et al., 1998).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In section 2 we introduce the graph transformation techniques used to projectivize and deprojectivize dependency graphs, and in section 3 we describe the data-driven dependency parser that is the core of our system.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Once the lexicon has been drawn, the model proceeds similarly to the standard token-level HMM: Emission parameters \u00ce\u00b8 are generated conditioned on tag assignments T . We also draw transition parameters \u00cf\u0086.","label":0},{"source_text":"There is no global pruning.","target_text":"Ignoring the identity of the target language words e and e0, the possible partial hypothesis extensions due to the IBM restrictions are shown in Table 2.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"MADA uses an ensemble of SVMs to first re-rank the output of a deterministic morphological analyzer.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"As such, global information from the whole context of a document is important to more accurately recognize named entities.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"So, we set a threshold that at least two examples are required to build a link.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Global context from the whole document is available and can be exploited in a natural manner with a maximum entropy classifier.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Another technique for parse hybridization is to use a na\u00efve Bayes classifier to determine which constituents to include in the parse.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"If e < b then the key is not found.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"A Single Generative Model for Joint Morphological Segmentation and Syntactic Parsing","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"There has been additional recent work on inducing lexicons or other knowledge sources from large corpora.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"When this feature type was included, CoBoost chose this default feature at an early iteration, thereby giving non-abstaining pseudo-labels for all examples, with eventual convergence to the two classifiers agreeing by assigning the same label to almost all examples.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"4.1 The Task and the Corpus.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Multiple features can be used for the same token.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Arabic is a morphologically rich language with a root-and-pattern system similar to other Semitic languages.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"5 We choose these two metrics over the Variation Information measure due to the deficiencies discussed in Gao and Johnson (2008).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We then show that the combining techniques presented above give better parsing accuracy than any of the individual parsers.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The question of what soft function to pick, and how to design' algorithms which optimize it, is an open question, but appears to be a promising way of looking at the problem.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Our derivation is slightly different from the one presented in (Schapire and Singer 98) as we restrict at to be positive.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Queries take the form p(wn|wn\u22121 1 ) where wn1 is an n-gram.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"This formulation of the constraint feature is equivalent to the use of a tagging dictionary extracted from the graph using a threshold T on the posterior distribution of tags for a given word type (Eq.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The relevance of the distinction between, say, phonological words and, say, dictionary words is shown by an example like rpftl_A :;!:Hfllil zhong1hua2 ren2min2 gong4he2-guo2 (China people republic) 'People's Republic of China.'","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Finally, we find links between sets of phrases, based on the NE instance pair data (for example, different phrases which link \u00e2\u0080\u009cIBM\u00e2\u0080\u009d and \u00e2\u0080\u009cLotus\u00e2\u0080\u009d) (Step 4).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"For each language and setting, we report one-to-one (11) and many- to-one (m-1) accuracies.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For a given \"word\" in the automatic segmentation, if at least k of the hu\u00c2\u00ad man judges agree that this is a word, then that word is considered to be correct.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The unlabeled data gives many such &quot;hints&quot; that two features should predict the same label, and these hints turn out to be surprisingly useful when building a classifier.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The binary language model from Section 5.2 and text phrase table were forced into disk cache before each run.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"2.6 Co-reference.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Output of the learning algorithm: a function h:Xxy [0, 1] where h(x, y) is an estimate of the conditional probability p(y1x) of seeing label y given that feature x is present.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Such analyzers propose multiple segmentation possibilities and their corresponding analyses for a token in isolation but have no means to determine the most likely ones.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In total, across all domains, we kept 13,976 phrases with keywords.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The number of latent HMM states for each language in our experiments was set to the number of fine tags in the language\u2019s treebank.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"18 77.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For every sequence of initial capitalized words, its longest substring that occurs in the same document as a sequence of initCaps is identified.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The token-level term is similar to the standard HMM sampling equations found in Johnson (2007).","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Table 5 shows the overall parsing accuracy attained with the three different encoding schemes, compared to the baseline (no special arc labels) and to training directly on non-projective dependency graphs.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"However, their inverted variant implements a reverse trie using less CPU and the same amount of memory7.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Affix Pron Base category N found N missed (recall) N correct (precision) t,-,7 The second issue is that rare family names can be responsible for overgeneration, especially if these names are otherwise common as single-hanzi words.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"4.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This causes a problem for reverse trie implementations, including SRILM itself, because it leaves n+1-grams without an n-gram node pointing to them.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"[Hasegawa et al. 04] reported only on relation discovery, but one could easily acquire para phrases from the results.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Given a sorted array A, these other packages use binary search to find keys in O(log |A|) time.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Besides size of training data, the use of dictionaries is another factor that might affect performance.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Yarowsky-cautious does not separate the spelling and contextual features, but does have a limit on the number of rules added at each stage.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"However, for our purposes it is not sufficient to repre\u00c2\u00ad sent the morphological decomposition of, say, plural nouns: we also need an estimate of the cost of the resulting word.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The local feature groups are: Non-Contextual Feature: This feature is set to 1 for all tokens.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"To establish a soft correspondence between the two languages, we use a second similarity function, which leverages standard unsupervised word alignment statistics (\u00a73.3).3 Since we have no labeled foreign data, our goal is to project syntactic information from the English side to the foreign side.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Given an anaphor and candidate, BABAR checks (1) whether the semantic classes of the anaphor intersect with the semantic expectations of the caseframe that extracts the candidate, and (2) whether the semantic classes of the candidate intersect with the semantic ex pectations of the caseframe that extracts the anaphor.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"t(i).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The model starts by generating a tag assignment for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"6 Conclusions.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Domain adaptation is a common concern when optimizing empirical NLP applications.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"95 B a s e li n e ( S e lf t a g ) 70 a l l B i k e l ( v 1 . 2 ) B a s e l i n e ( P r e t a g ) 7 0 a l l G o l d P O S 70 0.7 70 0.801 278 0.7 52 0.794 278 0.7 71 0.804 295 0.7 52 0.796 295 0.7 75 0.808 309 77.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"One of such approaches uses comparable documents, which are sets of documents whose content are found\/known to be almost the same, such as different newspaper stories about the same event [Shinyama and Sekine 03] or different translations of the same story [Barzilay 01].","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Experiments in two domains showed that the contextual role knowledge improved coreference performance, especially on pronouns.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Link phrases based on instance pairs Using NE instance pairs as a clue, we find links between sets of phrases.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We paid particular attention to minimize the number of free parameters, and used the same hyperparameters for all language pairs, rather than attempting language-specific tuning.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"A typical situation is shown in Figure 1.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"At each point during the derivation, the prediction is based on six word tokens, the two topmost tokens on the stack, and the next four input tokens.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","label":1},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"A novel element of our model is the ability to capture type-level tag frequencies.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"In this paper we have proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Our model outperforms theirs on four out of five languages on the best hyperparameter setting and three out of five on the median setting, yielding an average absolute difference across languages of 12.9% and 3.9% for best and median settings respectively compared to their best EM or LBFGS performance.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Here NO counts the number of hypothesized constituents in the development set that match the binary predicate specified as an argument.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"While the first three models get three to four tags wrong, our best model gets only one word wrong and is the most accurate among the four models for this example.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Hence, the resulting trees will no longer have two branches of equal size.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"2 60.7 56.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.2.4 Semantic Caseframe Expectations The third type of contextual role knowledge learned by BABAR is Semantic Caseframe Expectations.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Like in the co-reference annotation, G\u00c2\u00a8otze\u00e2\u0080\u0099s proposal has been applied by two annotators to the core corpus but it has not been systematically evaluated yet.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The probability distribution that satisfies the above property is the one with the highest entropy.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Along with locking and background kernel operations such as prefaulting, this explains why wall time is not one-eighth that of the single-threaded case. aLossy compression with the same weights. bLossy compression with retuned weights. the non-lossy options.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"4.4 Chinese Personal Names.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Linear weights are difficult to incorporate into the standard MERT procedure because they are \u201chidden\u201d within a top-level probability that represents the linear combination.1 Following previous work (Foster and Kuhn, 2007), we circumvent this problem by choosing weights to optimize corpus loglikelihood, which is roughly speaking the training criterion used by the LM and TM themselves.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The focus of this work is on building POS taggers for foreign languages, assuming that we have an English POS tagger and some parallel text between the two languages.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We hypothesize that modeling morphological information will greatly constrain the set of possible tags, thereby further refining the representation of the tag lexicon.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We can represent any derived tree of a TAG by the two substrings that appear in its frontier, and then define how the adjunction operation concatenates the substrings.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The semantic caseframe expectations are used in two ways.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Location list is processed into a list of unigrams and bigrams (e.g., New York).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Links can be of two different kinds: anaphoric or bridging (definite noun phrases picking up an antecedent via world-knowledge).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Furthermore, even the size of the dictionary per se is less important than the appropriateness of the lexicon to a particular test corpus: as Fung and Wu (1994) have shown, one can obtain substantially better segmentation by tailoring the lexicon to the corpus to be segmented.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In fact, during the first rounds many of the predictions of Th., g2 are zero.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"1 74.5 56.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Similarly, the tokens and are tested against each list, and if found, a corresponding feature will be set to 1.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We also performed experiments to evaluate the impact of each type of contextual role knowledge separately.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, co-occurring caseframes may reflect synonymy (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> abducted\u00e2\u0080\u009d) or related events (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> released\u00e2\u0080\u009d).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"These results are achieved by training on the official MUC6 and MUC7 training data, which is much less training data than is used by other machine learning systems that worked on the MUC6 or MUC7 named entity task (Bikel et al., 1997; Bikel et al., 1999; Borth- wick, 1999).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For queries, we uniformly sampled 10 million hits and 10 million misses.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Previous reports on Chinese segmentation have invariably cited performance either in terms of a single percent-correct score, or else a single precision-recall pair.","label":0},{"source_text":"Here we present two algorithms.","target_text":"A contextual rule considers words surrounding the string in the sentence in which it appears (e.g., a rule that any proper name modified by an appositive whose head is president is a person).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"During development, we sensed that the Recency and Syn- role KSs did not deserve to be on equal footing with the other KSs because their knowledge was so general.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"2.3 Rhetorical structure.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A dynamic programming algorithm is then used to select the sequence of word classes with the highest probability.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"2.2 Contextual Role Knowledge.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"It is known that CFG's, HG's, and TAG's can be recognized in polynomial time since polynomial time algorithms exist in for each of these formalisms.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"splitIN captures the verb\/preposition idioms that are widespread in Arabic.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We group the features used into feature groups.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"0 57.3 51.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We describe a novel approach for inducing unsupervised part-of-speech taggers for languages that have no labeled training data, but have translated text in a resource-rich language.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Whereas Arabic linguistic theory as Saddam (a) Reference (b) Stanford signs (1) and (2) to the class of pseudo verbs 01 +i J>1\u00ef\u00bf\u00bd inna and her sisters since they can beinflected, the ATB conventions treat (2) as a com plementizer, which means that it must be the head of SBAR.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"One of the difficulties in Natural Language Processing is the fact that there are many ways to express the same thing or event.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Note that in line 4 the last visited position for the successor hypothesis must be m. Otherwise , there will be four uncovered positions for the predecessor hypothesis violating the restriction.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Finally, the concatenated 5 * 20% output is used to train the reference resolution component.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For example, in .., says Mr. Cooper, a vice president of.. both a spelling feature (that the string contains Mr.) and a contextual feature (that president modifies the string) are strong indications that Mr. Cooper is of type Person.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"This offers the well-known advantages for inter- changability, but it raises the question of how to query the corpus across levels of annotation.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"of Articles No.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We do not attempt to identify the types of relationships that are found.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Figure 2 shows an excerpt of a sentence from the Italian test set and the tags assigned by four different models, as well as the gold tags.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Now, for this application one might be tempted to simply bypass the segmentation problem and pronounce the text character-by-character.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Until now, all evaluations of Arabic parsing\u00e2\u0080\u0094including the experiments in the previous section\u00e2\u0080\u0094have assumed gold segmentation.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The supervised POS tagging accuracies (on this tagset) are shown in the last row of Table 2.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Our method does not assume any knowledge about the target language (in particular no tagging dictionary is assumed), making it applicable to a wide array of resource-poor languages.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"we perform five runs with different random initialization of sampling state.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Another technique for parse hybridization is to use a na\u00efve Bayes classifier to determine which constituents to include in the parse.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Sequence of Initial Caps (SOIC): In the sentence Even News Broadcasting Corp., noted for its accurate reporting, made the erroneous announcement., a NER may mistake Even News Broadcasting Corp. as an organization name.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Figure 4 shows some such phrase sets based on keywords in the CC-domain.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"As is standard, we use a fixed constant K for the number of tagging states.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This highly effective approach is not directly applicable to the multinomial models used for core SMT components, which have no natural method for combining split features, so we rely on an instance-weighting approach (Jiang and Zhai, 2007) to downweight domain-specific examples in OUT.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The result of this is shown in Figure 7.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The combining algorithm is presented with the candidate parses and asked to choose which one is best.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"A novel element of our model is the ability to capture type-level tag frequencies.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Compared to related work on the recovery of long-distance dependencies in constituency-based parsing, our approach is similar to that of Dienes and Dubey (2003) in that the processing of non-local dependencies is partly integrated in the parsing process, via an extension of the set of syntactic categories, whereas most other approaches rely on postprocessing only.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"12 One class of full personal names that this characterization does not cover are married women's names.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This is a straightforward technique that is arguably better suited to the adaptation task than the standard method of treating representative IN sentences as queries, then pooling the match results.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"It can be easily shown from Thatcher's result that the path set of every local set is a regular set.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"It was then tested on section 22 of the Treebank in conjunction with the other parsers.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"There is a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099 of ten commentaries, for which the range of information (except for syntax) has been completed; the remaining data has been annotated to different degrees, as explained below.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"\u00e2\u0080\u00a2 Bridging links: the annotator is asked to specify the type as part-whole, cause-effect (e.g., She had an accident.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"1 53.8 47.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The transitive closure of the dictionary in (a) is composed with Id(input) (b) to form the WFST (c).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Microsoft\u2019s approach uses dependency trees, others use hierarchical phrase models.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"08 84.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The state in future has not enough work for its many teachers.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Our first algorithm is similar to Yarowsky's, but with some important modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Base NPs are the other significant category of nominal phrases.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Here, annotation proceeds in two phases: first, the domains and the units of IS are marked as such.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"An example of a fairly low-level relation is the affix relation, which holds between a stem morpheme and an affix morpheme, such as f1 -menD (PL).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We combined evidence from four contextual role knowledge sources with evidence from seven general knowledge sources using a DempsterShafer probabilistic model.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Overall, it gives improvements ranging from 1.1% for German to 14.7% for Italian, for an average improvement of 8.3% over the unsupervised feature-HMM model.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"See Section 5.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In general, different modalities (\u00e2\u0080\u009cplanned to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cagreed to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cbought\u00e2\u0080\u009d) were considered to express the same relationship within an extraction setting.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The number of paths that can be dependent is bounded by the grammar (in fact the maximum cardinality of a tree set determines this bound).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"If the same pair of NE instances is used with different phrases, these phrases are likely to be paraphrases.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Linear probing places at most one entry in each bucket.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"However, their system is a hybrid of hand-coded rules and machine learning methods.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"There is no global pruning.","target_text":"The computing time is low, since no reordering is carried out.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Linear probing places at most one entry in each bucket.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Two strings are letter equivalent if they contain equal number of occurrences of each terminal symbol, and two languages are letter equivalent if every string in one language is letter equivalent to a string in the other language and vice-versa.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In a few cases, the criteria for correctness are made more explicit.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For example: McCann initiated a new global system.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"In other words, the set of hidden states F was chosen to be the fine set of treebank tags.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Only tokens with initCaps not found in commonWords are tested against each list in Table 2.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The monotone search performs worst in terms of both error rates mWER and SSER.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"buy - acquire (5) buy - agree (2) buy - purchase (5) buy - acquisition (7) buy - pay (2)* buy - buyout (3) buy - bid (2) acquire - purchase (2) acquire - acquisition (2) acquire - pay (2)* purchase - acquisition (4) purchase - stake (2)* acquisition - stake (2)* unit - subsidiary (2) unit - parent (5) It is clear that these links form two clusters which are mostly correct.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"This leads to the best reported performance for robust non-projective parsing of Czech.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The semilinearity of Tree Adjoining Languages (TAL's), MCTAL's, and Head Languages (HL's) can be proved using this property, with suitable restrictions on the composition operations.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Two aspects of the corpus have been presented in previous papers ((Re- itter, Stede 2003) on underspecified rhetorical structure; (Stede 2003) on the perspective of knowledge-based summarization).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The correct resolution in sentence (b) comes from knowledge that people who are kidnapped are often subsequently released.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In many cases these failures in recall would be fixed by having better estimates of the actual prob\u00c2\u00ad abilities of single-hanzi words, since our estimates are often inflated.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Pseudo-labels are formed by taking seed labels on the labeled examples, and the output of the fixed classifier on the unlabeled examples.","label":0},{"source_text":"There is no global pruning.","target_text":"In the second and third translation examples, the IbmS word reordering performs worse than the QmS word reordering, since it can not take properly into account the word reordering due to the German verbgroup.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The table shows that the lexicon tag frequency predicated by our full model are the closest to the gold standard.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"However, there are several reasons why this approach will not in general work: 1.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Formalisms such as the restricted indexed grammars (Gazdar, 1985) and members of the hierarchy of grammatical systems given by Weir (1987) have independent paths, but more complex path sets.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"As a result, Habash et al.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Realizing gains in practice can be challenging, however, particularly when the target domain is distant from the background data.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Because we are working with only three parsers, the only situation in which context will help us is when it can indicate we should choose to believe a single parser that disagrees with the majority hypothesis instead of the majority hypothesis itself.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Proper names that match are resolved with each other.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In the third and final scheme, denoted Path, we keep the extra infor2Note that this is a baseline for the parsing experiment only (Experiment 2).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Virginia) and -sia are normally transliterated as fbSi!","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We formulate the update as follows: where \u2200ui \u2208 Vf \\ Vfl, \u03b3i(y) and \u03bai are defined as: We ran this procedure for 10 iterations.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In turn we use two sorts of heuristics, orthogonal to one another, to prune segmentation possibilities based on lexical and grammatical constraints.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"This distributional sparsity of syntactic tags is not unique to English 1 The source code for the work presented in this paper is available at http:\/\/groups.csail.mit.edu\/rbg\/code\/typetagging\/.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"It is not immediately obvious how to formulate an equivalent to equation (1) for an adapted TM, because there is no well-defined objective for learning TMs from parallel corpora.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The importance of dictionaries in NERs has been investigated in the literature (Mikheev et al., 1999).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The 2nd block contains the IR system, which was tuned by selecting text in multiples of the size of the EMEA training corpus, according to dev set performance.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Of these cases, 38 were temporal expressions (either a day of the week or month of the year).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"However, some caveats are in order in comparing this method (or any method) with other approaches to seg\u00c2\u00ad mentation reported in the literature.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For previously unseen hanzi in given names, Chang et al. assign a uniform small cost; but we know that some unseen hanzi are merely acci\u00c2\u00ad dentally missing, whereas others are missing for a reason-for example, because they have a bad connotation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"i..f,..","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Since we could not bias the subjects towards a particular segmentation and did not presume linguistic sophistication on their part, the instructions were simple: subjects were to mark all places they might plausibly pause if they were reading the text aloud.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In the current work morphological analyses and lexical probabilities are derived from a small Treebank, which is by no means the best way to go.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Moving beyond directly related work, major themes in SMT adaptation include the IR (Hildebrand et al., 2005; L\u00a8u et al., 2007; Zhao et al., 2004) and mixture (Finch and Sumita, 2008; Foster and Kuhn, 2007; Koehn and Schroeder, 2007; L\u00a8u et al., 2007) approaches for LMs and TMs described above, as well as methods for exploiting monolingual in-domain text, typically by translating it automatically and then performing self training (Bertoldi and Federico, 2009; Ueffing et al., 2007; Schwenk and Senellart, 2009).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Fig.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We often encounter sentences that are highly ambiguous in themselves, without some prior knowledge of the entities concerned.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Finally, we show that in application settings, the absence of gold segmentation lowers parsing performance by 2\u00e2\u0080\u00935% F1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(a) ;IE shi4 'be' => ;IE;IE shi4bu2-shi4 (be-not-be) 'is it?'","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"3 58.3 40.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Having found (spelling, context) pairs in the parsed data, a number of features are extracted.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The implementation is In a similar effort, (G\u00c2\u00a8otze 2003) developed a proposal for the theory-neutral annotation of information structure (IS) \u00e2\u0080\u0094 a notoriously difficult area with plenty of conflicting and overlapping terminological conceptions.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"63 81.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"i..f,..","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We used the TF\/ITF metric to identify keywords.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Surprisingly, the non-parametric switching technique also exhibited robust behaviour in this situation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Third, all remaining anaphora are evaluated by 11 different knowledge sources: the four contextual role knowledge sources just described and seven general knowledge sources.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The paper is organized as follows: Section 2 explains the different layers of annotation that have been produced or are being produced.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Unfortunately, there is no standard corpus of Chinese texts, tagged with either single or multiple human judgments, with which one can compare performance of various methods.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Clearly, for judges h and h taking h as standard and computing the precision and recall for Jz yields the same results as taking h as the standard, and computing for h, 14 All evaluation materials, with the exception of those used for evaluating personal names were drawn.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Table 3 Classes of words found by ST for the test corpus.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kind of cue.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Find keywords for each NE pair When we look at the contexts for each domain, we noticed that there is one or a few important words which indicate the relation between the NEs (for example, the word \u00e2\u0080\u009cunit\u00e2\u0080\u009d for the phrase \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Unique Occurrences and Zone (UNIQ): This group of features indicates whether the word is unique in the whole document.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Following the setup of Johnson (2007), we use the whole of the Penn Treebank corpus for training and evaluation on English.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Two measures that can be used to compare judgments are: 1.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"While RST (Mann, Thompson 1988) proposed that a single relation hold between adjacent text segments, SDRT (Asher, Lascarides 2003) maintains that multiple relations may hold simultaneously.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Experiments in two domains showed that the contextual role knowledge improved coreference performance, especially on pronouns.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A Stochastic Finite-State Word-Segmentation Algorithm for Chinese","label":1},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Among the IS-units, the referring expressions are marked as such and will in the second phase receive a label for cognitive status (active, accessible- text, accessible-situation, inferrable, inactive).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"6.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Nonetheless there is no alternative to Reiche\u00e2\u0080\u0099s plan.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"5.1 Parsing Models.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The model was parameterized such that the joint probability of a (label, feature-set) pair P(yi, xi) is written as The model assumes that (y, x) pairs are generated by an underlying process where the label is first chosen with some prior probability P(yi); the number of features mi is then chosen with some probability P(mi); finally the features are independently generated with probabilities P(xulyi).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The 13 positions of the source sentence are processed in the order shown.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"In general, several gross corpus statistics favor the ATB, so other factors must contribute to parsing underperformance.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Given a document to process, BABAR uses four modules to perform coreference resolution.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"While size of the resulting transducers may seem daunting-the segmenter described here, as it is used in the Bell Labs Mandarin TTS system has about 32,000 states and 209,000 arcs-recent work on minimization of weighted machines and transducers (cf.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"basically complete, yet some improvements and extensions are still under way.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"With regard to purely morphological phenomena, certain processes are not han\u00c2\u00ad dled elegantly within the current framework Any process involving reduplication, for instance, does not lend itself to modeling by finite-state techniques, since there is no way that finite-state networks can directly implement the copying operations required.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Note that while the standard HMM, has O(K n) emission parameters, our model has O(n) effective parameters.3 Token Component Once HMM parameters (\u00cf\u0086, \u00ce\u00b8) have been drawn, the HMM generates a token-level corpus w in the standard way: P (w, t|\u00cf\u0086, \u00ce\u00b8) = P (T , W , \u00ce\u00b8, \u00cf\u0088, \u00cf\u0086, t, w|\u00ce\u00b1, \u00ce\u00b2) = P (T , W , \u00cf\u0088|\u00ce\u00b2) [Lexicon] \u00ef\u00a3\u00ab n n \u00ef\u00a3\u00ad (w,t)\u00e2\u0088\u0088(w,t) j \u00ef\u00a3\u00b6 P (tj |\u00cf\u0086tj\u00e2\u0088\u00921 )P (wj |tj , \u00ce\u00b8tj )\u00ef\u00a3\u00b8 P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) [Parameter] P (w, t|\u00cf\u0086, \u00ce\u00b8) [Token] We refer to the components on the right hand side as the lexicon, parameter, and token component respectively.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Future work along these lines will incorporate other layers of annotation, in particular the syntax information.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"One is the accuracy within a set of phrases which share the same keyword; the other is the accuracy of links.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We will report the evaluation results in the next subsection.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"nan2gual 'pumpkin.'","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Graph construction does not require any labeled data, but makes use of two similarity functions.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A Stochastic Finite-State Word-Segmentation Algorithm for Chinese","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In IdentiFinder, there is a priority in the feature assignment, such that if one feature is used for a token, another feature lower in priority will not be used.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Although this feature helps, we encounter one consequence of variable word order.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"They return a value in the range [0,1], where 0 indicates neutrality and 1 indicates the strongest belief that the candidate and anaphor are coreferent.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"This model admits a simple Gibbs sampling algorithm where the number of latent variables is proportional to the number of word types, rather than the size of a corpus as for a standard HMM sampler (Johnson, 2007).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"with the number of exactly matching guess trees.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"In the label propagation stage, we propagate the automatic English tags to the aligned Italian trigram types, followed by further propagation solely among the Italian vertices. the Italian vertices are connected to an automatically labeled English vertex.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"In our experiments, we used the same set of features as BergKirkpatrick et al. (2010): an indicator feature based In a traditional Markov model, the emission distribution P\u0398(Xi = xi  |Zi = zi) is a set of multinomials.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Virginia) and -sia are normally transliterated as fbSi!","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Section 2.2 then describes our representation for contextual roles and four types of contextual role knowledge that are learned from the training examples.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"98 15.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Central to our approach (see Algorithm 1) is a bilingual similarity graph built from a sentence-aligned parallel corpus.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The token-level term is similar to the standard HMM sampling equations found in Johnson (2007).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Maamouri et al.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"As the name implies, space is O(m) and linear in the number of entries.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Language models that contain wi must also contain prefixes wi for 1 G i G k. Therefore, when the model is queried for p(wnjwn\u22121 1 ) but the longest matching suffix is wnf , it may return state s(wn1) = wnf since no longer context will be found.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"A Person-Prefix-List is compiled in an analogous way.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Note that on some examples (around 2% of the test set) CoBoost abstained altogether; in these cases we labeled the test example with the baseline, organization, label.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"More examples are shown in Figure 5.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"lla\/llb and 14a\/14b respectively).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"So, who won the competition?","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The second main result is that the pseudo-projective approach to parsing (using special arc labels to guide an inverse transformation) gives a further improvement of about one percentage point on attachment score.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In addition to the automatic methods, AG, GR, and ST, just discussed, we also added to the plot the values for the current algorithm using only dictionary entries (i.e., no productively derived words or names).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"m(S) represents the belief that the correct hypothesis is included in S. The model assumes that evidence also arrives as a probability density function (pdf) over sets of hypotheses.6 Integrating new evidence into the existing model is therefore simply a matter of defining a function to merge pdfs, one representing the current belief system and one representing the beliefs of the new evidence.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The two new terms force the two classifiers to agree, as much as possible, on the unlabeled examples.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Output of the learning algorithm: a function h:Xxy [0, 1] where h(x, y) is an estimate of the conditional probability p(y1x) of seeing label y given that feature x is present.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We trained this model by optimizing the following objective function: Note that this involves marginalizing out all possible state configurations z for a sentence x, resulting in a non-convex objective.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The average agreement among the human judges is .76, and the average agreement between ST and the humans is .75, or about 99% of the interhuman agreement.15 One can better visualize the precision-recall similarity matrix by producing from that matrix a distance matrix, computing a classical metric multidimensional scaling (Torgerson 1958; Becker, Chambers, Wilks 1988) on that dis\u00c2\u00ad tance matrix, and plotting the first two most significant dimensions.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We suspect that the higher precision in the disasters domain may be due to its substantially larger training corpus.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The maximum likelihood estimates (i.e., parameter values which maximize 10) can not be found analytically, but the EM algorithm can be used to hill-climb to a local maximum of the likelihood function from some initial parameter settings.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Two of the Mainlanders also cluster close together but, interestingly, not particularly close to the Taiwan speakers; the third Mainlander is much more similar to the Taiwan speakers.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We use N(u) to denote the neighborhood of vertex u, and fixed n = 5 in our experiments.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We further thank Khalil Simaan (ILLCUvA) for his careful advise concerning the formal details of the proposal.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The above auxiliary quantity satisfies the following recursive DP equation: Qe0 (e; C; j) = Initial Skip Verb Final 1.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The second term is a regularizer and encourages all type marginals to be uniform to the extent that is allowed by the first two terms (cf. maximum entropy principle).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For MUC7, there are also no published results on systems trained on only the official training data of 200 aviation disaster articles.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This is appropriate in cases where it is sanctioned by Bayes\u2019 law, such as multiplying LM and TM probabilities, but for adaptation a more suitable framework is often a mixture model in which each event may be generated from some domain.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"These enable much larger models in memory, compensating for lost accuracy.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Can we do . QmS: Yes, wonderful.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"A different but supplementary perspective on discourse-based information structure is taken 11ventionalized patterns (e.g., order of informa by one of our partner projects, which is inter tion in news reports).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Finally, the concatenated 5 * 20% output is used to train the reference resolution component.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"A corpus of German newspaper commentaries has been assembled and annotated with different information (and currently, to different degrees): part-of-speech, syntax, rhetorical structure, connectives, co-reference, and information structure.","label":1},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Like in the co-reference annotation, G\u00c2\u00a8otze\u00e2\u0080\u0099s proposal has been applied by two annotators to the core corpus but it has not been systematically evaluated yet.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The work reported here is closely related to [Ha- segawa et al. 04].","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(Specifically, the limit n starts at 5 and increases by 5 at each iteration.)","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We allow any number of bits from 2 to 25, unlike IRSTLM (8 bits) and BerkeleyLM (17\u221220 bits).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The natural baseline approach is to concatenate data from IN and OUT.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The confidence level is then used as the belief value for the knowledge source.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"There are thus some very good reasons why segmentation into words is an important task.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"This process produces a large set of caseframes coupled with a list of the noun phrases that they extracted.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"5 64.7 42.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"(2009) also report results on English, but on the reduced 17 tag set, which is not comparable to ours).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Step 3.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The compressed variant uses block compression and is rather slow as a result.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"It is not easy to make a clear definition of \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For example: McCann initiated a new global system.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"It has been shown for English (Wang and Hirschberg 1992; Hirschberg 1993; Sproat 1994, inter alia) that grammatical part of speech provides useful information for these tasks.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"13.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We considered using the MUC6 and MUC7 data sets, but their training sets were far too small to learn reliable co-occurrence statistics for a large set of contextual role relationships.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"However there is no global pruning.","label":0},{"source_text":"There is no global pruning.","target_text":"Table 4 shows translation results for the three approaches.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Its success depends on the two domains being relatively close, and on the OUT corpus not being so large as to overwhelm the contribution of IN.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For each label (Per s on, organization and Location), take the n contextual rules with the highest value of Count' (x) whose unsmoothed3 strength is above some threshold pmin.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Steedman (1986) considers Categorial Grammars in which both the operations of function application and composition may be used, and in which function can specify whether they take their arguments from their right or left.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We considered using the MUC6 and MUC7 data sets, but their training sets were far too small to learn reliable co-occurrence statistics for a large set of contextual role relationships.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The prediction based on these features is a knearest neighbor classification, using the IB1 algorithm and k = 5, the modified value difference metric (MVDM) and class voting with inverse distance weighting, as implemented in the TiMBL software package (Daelemans et al., 2003).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Ex: The brigade, which attacked ...","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"A named entity recognizer (NER) is useful in many NLP applications such as information extraction, question answering, etc. On its own, a NER can also provide users who are looking for person or organization names with quick information.","label":1},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The 13 positions of the source sentence are processed in the order shown.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Our use of an unweighted lattice reflects our belief that all the segmentations of the given input sentence are a-priori equally likely; the only reason to prefer one segmentation over the another is due to the overall syntactic context which is modeled via the PCFG derivations.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"na me =>2 ha nzi fa mi ly 2 ha nzi gi ve n 5.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Aside from adding a simple rule to correct alif deletion caused by the preposition J, no other language-specific processing is performed.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The pre terminal morphological analyses are mapped to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags provided with the tree- bank.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Training examples are generated automatically by identifying noun phrases that can be easily resolved with their antecedents using lexical and syntactic heuristics.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Morphologically derived words such as, xue2shengl+men0.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"If no candidate satisfies this condition (which is often the case), then the anaphor is left unresolved.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"level.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"To provide a thorough analysis, we evaluated three baselines and two oracles in addition to two variants of our graph-based approach.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Recently, combination techniques have been investigated for part of speech tagging with positive results (van Halteren et al., 1998; Brill and Wu, 1998).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"M works in space S(n) if for every string that M accepts no configuration exceeds space S(n).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Nonstochastic lexical-knowledge-based approaches have been much more numer\u00c2\u00ad ous.","label":0},{"source_text":"There is no global pruning.","target_text":"The following auxiliary quantity is defined: Qe0 (e; C; j) := probability of the best partial hypothesis (ei 1; bi 1), where C = fbkjk = 1; ; ig, bi = j, ei = e and ei\u00f4\u0080\u0080\u00801 = e0.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Asking the annotator to also formulate the question is a way of arriving at more reproducible decisions.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"We use MMAX for this annotation as well.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(3)) to be defined over unlabeled as well as labeled instances.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In Chinese text, individual characters of the script, to which we shall refer by their traditional name of hanzi,Z are written one after another with no intervening spaces; a Chinese sentence is shown in Figure 1.3 Partly as a result of this, the notion \"word\" has never played a role in Chinese philological tradition, and the idea that Chinese lacks any\u00c2\u00ad thing analogous to words in European languages has been prevalent among Western sinologists; see DeFrancis (1984).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The alignment mapping is j ! i = aj from source position j to target position i = aj . The use of this alignment model raises major problems if a source word has to be aligned to several target words, e.g. when translating German compound nouns.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"3.3 Evaluation Results.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The pseudo-code describing the algorithm is given in Fig.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Process nominals name the action of the transitive or ditransitive verb from which they derive.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Hence we decided to restrict ourselves to only information from the same document.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"(Webber et al., 2003)).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The \u201cNo LP\u201d model does not outperform direct projection for German and Greek, but performs better for six out of eight languages.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"That is, we can use the discourse parser on PCC texts, emulating for instance a \u00e2\u0080\u009cco-reference oracle\u00e2\u0080\u009d that adds the information from our co-reference annotations.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Thus it is possible, for illustration, to look for a noun phrase (syntax tier) marked as topic (information structure tier) that is in a bridging relation (co-reference tier) to some other noun phrase.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"C is the union of the sets of constituents suggested by the parsers. r(c) is a binary function returning t (for true) precisely when the constituent c E C should be included in the hypothesis.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.2.3 Lexical Caseframe Expectations The second type of contextual role knowledge learned by BABAR is Lexical Caseframe Expectations, which are used by the CFLex knowledge source.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The AdaBoost algorithm was developed for supervised learning.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"5 64.7 42.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"A dynamic programming recursion similar to the one in Eq. 2 is evaluated.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"One way to approach this discrepancy is to assume a preceding phase of morphological segmentation for extracting the different lexical items that exist at the token level (as is done, to the best of our knowledge, in all parsing related work on Arabic and its dialects (Chiang et al., 2006)).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We also thank ChaoHuang Chang, reviewers for the 1994 ACL conference, and four anonymous reviewers for Computational Linguistics for useful comments.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"0 D\/ nc 5.0 The minimal dictionary encoding this information is represented by the WFST in Figure 2(a).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In total, across all domains, we kept 13,976 phrases with keywords.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Our experiments all concern the analytical annotation, and the first experiment is based only on the training part.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"All language model queries issued by machine translation decoders follow a left-to-right pattern, starting with either the begin of sentence token or null context for mid-sentence fragments.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Unification is used in LFG's to produce structures having two dependent spines of unbounded length as in Figure 5.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"On the other hand, we are interested in the application of rhetorical analysis or \u00e2\u0080\u0098discourse parsing\u00e2\u0080\u0099 (3.2 and 3.3), in text generation (3.4), and in exploiting the corpus for the development of improved models of discourse structure (3.5).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"While RST (Mann, Thompson 1988) proposed that a single relation hold between adjacent text segments, SDRT (Asher, Lascarides 2003) maintains that multiple relations may hold simultaneously.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Much of the study of grammatical systems in computational linguistics has been focused on the weak generative capacity of grammatical formalism.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Features and context were initially introduced into the models, but they refused to offer any gains in performance.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"7.96 5.55 1 l...................................................................................................................................................................................................J..","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Corporate Suffixes and Person Prefixes of Other Occurrences (CSPP): If McCann has been seen as Mr. McCann somewhere else in the document, then one would like to give person a higher probability than organization.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Thus we have some confidence that our own performance is at least as good as that of Chang et al.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Church and Hanks [1989]), and we have used lists of character pairs ranked by mutual information to expand our own dictionary.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Table 9: Dev set results for sentences of length \u00e2\u0089\u00a4 70.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Note that the sets of possible classifiers for a given noun can easily be encoded on that noun by grammatical features, which can be referred to by finite-state grammatical rules.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"On the other hand, only very restricted reorderings are necessary, e.g. for the translation direction from Table 2: Coverage set hypothesis extensions for the IBM reordering.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Let us notate the set of previously unseen, or novel, members of a category X as unseen(X); thus, novel members of the set of words derived in f, menO will be de\u00c2\u00ad noted unseen(f,).","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The most informative scheme, Head+Path, gives the highest scores, although with respect to Head the difference is not statistically significant, while the least informative scheme, Path \u2013 with almost the same performance on treebank transformation \u2013 is significantly lower (p < 0.01).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We run the baseline Moses system for the French-English track of the 2011 Workshop on Machine Translation,9 translating the 3003-sentence test set.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The two new terms force the two classifiers to agree, as much as possible, on the unlabeled examples.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The independence of paths in the tree sets of the k tI grammatical formalism in this hierarchy can be shown by means of tree pumping lemma of the form t1ti3t .","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Global features are extracted from other occurrences of the same token in the whole document.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"orthographic words are thus only a starting point for further analysis and can only be regarded as a useful hint at the desired division of the sentence into words.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Again, this deserves further investigation.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Relative pronouns with only 1 NP in scope..","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"98 15.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The foreign language vertices (denoted by Vf) correspond to foreign trigram types, exactly as in Subramanya et al. (2010).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The development of a na\u00efve Bayes classifier involves learning how much each parser should be trusted for the decisions it makes.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Denote the unthresholded classifiers after t \u2014 1 rounds by git\u20141 and assume that it is the turn for the first classifier to be updated while the second one is kept fixed.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Performance typically stabilizes across languages after only a few number of iterations.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Given a sorted array A, these other packages use binary search to find keys in O(log |A|) time.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Jud ges A G G R ST M 1 M 2 M 3 T1 T2 T3 AG 0.7 0 0.7 0 0 . 4 3 0.4 2 0.6 0 0.6 0 0.6 2 0.5 9 GR 0.9 9 0 . 6 2 0.6 4 0.7 9 0.8 2 0.8 1 0.7 2 ST 0 . 6 4 0.6 7 0.8 0 0.8 4 0.8 2 0.7 4 M1 0.7 7 0.6 9 0.7 1 0.6 9 0.7 0 M2 0.7 2 0.7 3 0.7 1 0.7 0 M3 0.8 9 0.8 7 0.8 0 T1 0.8 8 0.8 2 T2 0.7 8 respectively, the recall and precision.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This method, one instance of which we term the \"greedy algorithm\" in our evaluation of our own system in Section 5, involves starting at the beginning (or end) of the sentence, finding the longest word starting (ending) at that point, and then repeating the process starting at the next (previous) hanzi until the end (begin\u00c2\u00ad ning) of the sentence is reached.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The results, along with the total number of phrases, are shown in Table 1.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In addition to a heuristic based on decision list learning, we also presented a boosting-like framework that builds on ideas from (Blum and Mitchell 98).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"37 79.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(b) 89 :1 t& tal de cai2neng2 hen3 he DE talent very 'He has great talent' f.b ga ol hig h While the current algorithm correctly handles the (b) sentences, it fails to handle the (a) sentences, since it does not have enough information to know not to group the sequences.ma3lu4 and?]cai2neng2 respectively.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"2.1 Part-of-speech tags.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"For example, CFG's cannot produce trees of the form shown in Figure 1 in which there are nested dependencies between S and NP nodes appearing on the spine of the tree.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"At present, the \u00e2\u0080\u0098Potsdam Commentary Corpus\u00e2\u0080\u0099 (henceforth \u00e2\u0080\u0098PCC\u00e2\u0080\u0099 for short) consists of 170 commentaries from Ma\u00c2\u00a8rkische Allgemeine Zeitung, a German regional daily.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"He used an additional maximum entropy classifier that tries to correct mistakes by using reference resolution.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"By this assumption, each element x E X can also be represented as (xi, x2) E X1 x X2.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Gan's solution depends upon a fairly sophisticated language model that attempts to find valid syntactic, semantic, and lexical relations between objects of various linguistic types (hanzi, words, phrases).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Third, we develop a human interpretable grammar that is competitive with a latent variable PCFG.","label":0},{"source_text":"There is no global pruning.","target_text":"While processing the source sentence monotonically, the initial state I is entered whenever there are no uncovered positions to the left of the rightmost covered position.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The matching sentence pairs are then added to the IN corpus, and the system is re-trained.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We showed in \u00c2\u00a72 that lexical ambiguity explains the underperformance of these categories.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Much confusion has been sown about Chinese writing by the use of the term ideograph, suggesting that hanzi somehow directly represent ideas.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Its success depends on the two domains being relatively close, and on the OUT corpus not being so large as to overwhelm the contribution of IN.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Quantization can be improved by jointly encoding probability and backoff.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Less frequently studied is the interplay among language, annotation choices, and parsing model design (Levy and Manning, 2003; Ku\u00c2\u00a8 bler, 2005).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"It is generally considered to be fast (Pauls 29 \u2212 1 probabilities and 2' \u2212 2 non-zero backoffs. and Klein, 2011), with a default implementation based on hash tables within each trie node.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Table 5 shows the overall parsing accuracy attained with the three different encoding schemes, compared to the baseline (no special arc labels) and to training directly on non-projective dependency graphs.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Each trie node is individually allocated and full 64-bit pointers are used to find them, wasting memory.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Bikel et al.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"These methods demonstrated the benefits of incorporating linguistic features using a log-linear parameterization, but requires elaborate machinery for training.","label":0},{"source_text":"Here we present two algorithms.","target_text":"They also describe an application of cotraining to classifying web pages (the to feature sets are the words on the page, and other pages pointing to the page).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thus, as in CFG's, at any point in the derivation, the set of structures that can be applied is determined only by a finite set of rules encapsulated by the grammar.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"41.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"\u00e2\u0080\u00a2 Anaphoric links: the annotator is asked to specify whether the anaphor is a repetition, partial repetition, pronoun, epithet (e.g., Andy Warhol \u00e2\u0080\u0093 the PopArt artist), or is-a (e.g., Andy Warhol was often hunted by photographers.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Part-of-speech (POS) tag distributions are known to exhibit sparsity \u00e2\u0080\u0094 a word is likely to take a single predominant tag in a corpus.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Also, \u00e2\u0080\u009cagree\u00e2\u0080\u009d in the CC-domain is not a desirable keyword.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Personal names such as 00, 3R; zhoulenl-lai2 'Zhou Enlai.'","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"If an unlabeled vertex does not have a path to any labeled vertex, this term ensures that the converged marginal for this vertex will be uniform over all tags, allowing the middle word of such an unlabeled vertex to take on any of the possible tags.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In this example there are four \"input characters,\" A, B, C and D, and these map respectively to four \"pronunciations\" a, b, c and d. Furthermore, there are four \"words\" represented in the dictionary.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The fraction of buckets that are empty is m\u22121 m , so average lookup time is O( m 1) and, crucially, constant in the number of entries.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Note that these observa sider suffix features, capitalization features, punctuation, and digit features.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"72 77.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Suppose M has to determine whether the k substrings ,.. .,ak can be derived from some symbol A.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The results, along with the total number of phrases, are shown in Table 1.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For the experiments in this paper we made a couple of additional modifications to the CoBoost algorithm.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Experiments using data from the Prague Dependency Treebank show that the combined system can handle nonprojective constructions with a precision sufficient to yield a significant improvement in overall parsing accuracy.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Part-of-speech (POS) tag distributions are known to exhibit sparsity \u00e2\u0080\u0094 a word is likely to take a single predominant tag in a corpus.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"However, there is a crucial difference: the morphological probabilities in their model come from discriminative models based on linear context.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"That is, we can use the discourse parser on PCC texts, emulating for instance a \u00e2\u0080\u009cco-reference oracle\u00e2\u0080\u009d that adds the information from our co-reference annotations.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"In all of our experiments, the binary file (whether mapped or, in the case of most other packages, interpreted) is loaded into the disk cache in advance so that lazy mapping will never fault to disk.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For each token , zero, one, or more of the features in each feature group are set to 1.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"For the experiments, we used four newswire corpora, the Los Angeles Times\/Washington Post, The New York Times, Reuters and the Wall Street Journal, all published in 1995.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Second, we identified the 100 most frequent nouns in the training corpus and manually labeled them with semantic tags.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"97 78.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"G1 and G2 are hanzi, we can estimate the probability of the sequence being a name as the product of: \u00e2\u0080\u00a2 the probability that a word chosen randomly from a text will be a name-p(rule 1), and \u00e2\u0080\u00a2 the probability that the name is of the form 1hanzi-family 2hanzi-given-p(rule 2), and \u00e2\u0080\u00a2 the probability that the family name is the particular hanzi F1-p(rule 6), and \u00e2\u0080\u00a2 the probability that the given name consists of the particular hanzi G1 and G2-p(rule 9) This model is essentially the one proposed in Chang et al.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"(Thus the domain of the dev and test corpora matches IN.)","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The natural language processing community is in the strong position of having many available approaches to solving some of its most fundamental problems.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We have argued that the proposed method performs well.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Otherwise, label the training data with the combined spelling\/contextual decision list, then induce a final decision list from the labeled examples where all rules (regardless of strength) are added to the decision list.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Non-empty buckets contain an entry belonging to them or to a preceding bucket where a conflict occurred.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In contrast, even though context can be helpful for resolving definite NPs, context can be trumped by the semantics of the nouns themselves.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"If somewhere else in the document we see \u00e2\u0080\u009crestrictions put in place by President Bush\u00e2\u0080\u009d, then we can be surer that Bush is a name.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The results, along with the total number of phrases, are shown in Table 1.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"This scenario is applicable to a large set of languages and has been considered by a number of authors in the past (Alshawi et al., 2000; Xi and Hwa, 2005; Ganchev et al., 2009).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"2) An improved language model, which takes into account syntactic structure, e.g. to ensure that a proper English verbgroup is generated.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"4.2 Global Features.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Let us notate the set of previously unseen, or novel, members of a category X as unseen(X); thus, novel members of the set of words derived in f, menO will be de\u00c2\u00ad noted unseen(f,).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our assumption is that caseframes that co-occur in resolutions often have a 2 This normalization is performed syntactically without semantics, so the agent and patient roles are not guaranteed to hold, but they usually do in practice.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Sometime, multiple words are needed, like \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, \u00e2\u0080\u009cprime minister\u00e2\u0080\u009d or \u00e2\u0080\u009cpay for\u00e2\u0080\u009d (\u00e2\u0080\u009cpay\u00e2\u0080\u009d and \u00e2\u0080\u009cpay for\u00e2\u0080\u009d are different senses in the CC-domain).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Feature weights were set using Och\u2019s MERT algorithm (Och, 2003).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The second row is the accuracy of the best of the three parsers.'","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This logic applies recursively: if wnf+1 similarly does not extend and has zero log backoff, it too should be omitted, terminating with a possibly empty context.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This is most severe with RandLM in the multi-threaded case, where each thread keeps a separate cache, exceeding the original model size.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Two issues distinguish the various proposals.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Although matching is done at the sentence level, this information is subsequently discarded when all matches are pooled.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Note that these observa sider suffix features, capitalization features, punctuation, and digit features.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"If the log backoff of wnf is also zero (it may not be in filtered models), then wf should be omitted from the state.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Different sentence structure and rich target language morphology are two reasons for this.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The procedure using the tagged sentences to discover paraphrases takes about one hour on a 2GHz Pentium 4 PC with 1GB of memory.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Intuitively, it places more weight on OUT when less evidence from IN is available.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"4.1 Dictionary Representation.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"In both cases, SRILM walks its trie an additional time to minimize context as mentioned in Section 4.1.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Another option is the closedsource data structures from Sheffield (Guthrie and Hepple, 2010).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"When 7' is adjoined at ?I in the tree 7 we obtain a tree v&quot;.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Exposing this information to the decoder will lead to better hypothesis recombination.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"On several languages, we report performance exceeding that of state-of-the art systems.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Our first algorithm is similar to Yarowsky's, but with some important modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This means that the PCC cannot grow particularly quickly.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Statistical methods seem particularly applicable to the problem of unknown-word identification, especially for constructions like names, where the linguistic constraints are minimal, and where one therefore wants to know not only that a particular se\u00c2\u00ad quence of hanzi might be a name, but that it is likely to be a name with some probabil\u00c2\u00ad ity.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For each extension a new position is added to the coverage set.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"A detailed description of the search procedure used is given in this patent.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The ATB gives several different analyses to these words to indicate different types of coordination.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This task measures how well each package performs in machine translation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The performance of our system on those sentences ap\u00c2\u00ad peared rather better than theirs.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Given a document to process, BABAR uses four modules to perform coreference resolution.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"07 80.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Using the concept of inverted alignments, we explicitly take care of the coverage constraint by introducing a coverage set C of source sentence positions that have been already processed.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"These knowledge sources determine whether the contexts surrounding an anaphor and antecedent are compatible.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Feature-based HMM Model (Berg- Kirkpatrick et al., 2010): The KM model uses a variety of orthographic features and employs the EM or LBFGS optimization algorithm; Posterior regulariation model (Grac\u00c2\u00b8a et al., 2009): The G10 model uses the posterior regular- ization approach to ensure tag sparsity constraint.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The joint morphological and syntactic hypothesis was first discussed in (Tsarfaty, 2006; Tsarfaty and Sima\u2019an, 2004) and empirically explored in (Tsarfaty, 2006).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"There is no global pruning.","target_text":"Subsets of partial hypotheses with coverage sets C of increasing cardinality c are processed.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"IRSTLM 5.60.02 (Federico et al., 2008) is a sorted trie implementation designed for lower memory consumption.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"logical rules, and personal names; the transitive closure of the resulting machine is then computed.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The first m pairs have labels yi, whereas for i = m + 1, , n the pairs are unlabeled.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"First we present the non-parametric version of parser switching, similarity switching: The intuition for this technique is that we can measure a similarity between parses by counting the constituents they have in common.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"4.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Of course, since the number of attested (phonemic) Mandarin syllables (roughly 1400, including tonal distinctions) is far smaller than the number of morphemes, it follows that a given syllable could in principle be written with any of several different hanzi, depending upon which morpheme is intended: the syllable zhongl could be lfl 'middle,''clock,''end,' or ,'loyal.'","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We refer to a segment and its assigned PoS tag as a lexeme, and so analyses are in fact sequences of lexemes.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The confidence intervals are computed by bootstrap resampling for BLEU, and by standard significance testing for the manual scores, as described earlier in the paper.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Moses sets the cache size parameter to 50 so we did as well; the resulting cache size is 2.82 GB.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The pre terminal morphological analyses are mapped to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags provided with the tree- bank.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We evaluate our approach on seven languages: English, Danish, Dutch, German, Portuguese, Spanish, and Swedish.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For corporate suffixes, a list of tokens cslist that occur frequently as the last token of an organization name is collected from the training data.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"KenLM: Faster and Smaller Language Model Queries","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"There are two weaknesses in Chang et al.'s model, which we improve upon.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"In our work, we demonstrate that using a simple na\u00c2\u00a8\u00c4\u00b1veBayes approach also yields substantial performance gains, without the associated training complexity.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The matching sentence pairs are then added to the IN corpus, and the system is re-trained.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Finally, we thank Kuzman Ganchev and the three anonymous reviewers for helpful suggestions and comments on earlier drafts of this paper.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The path set of a tree set is the union of the path sets of trees in that tree set.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"we perform five runs with different random initialization of sampling state.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Along with locking and background kernel operations such as prefaulting, this explains why wall time is not one-eighth that of the single-threaded case. aLossy compression with the same weights. bLossy compression with retuned weights. the non-lossy options.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Let H be the set of hanzi, p be the set of pinyin syllables with tone marks, and P be the set of grammatical part-of-speech labels.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR applies a DempsterShafer probabilistic model to make resolutions based on evidence from the contextual role knowledge sources as well as general knowledge sources.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Table 1 shows results of the benchmark.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"A very small excerpt from an Italian-English graph is shown in Figure 1.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"With the exception of the Dutch data set, no other processing is performed on the annotated tags.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"To differentiate between the coordinating and discourse separator functions of conjunctions (Table 3), we mark each CC with the label of its right sister (splitCC).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Particular relations are also consistent with particular hypotheses about the segmentation of a given sentence, and the scores for particular relations can be incremented or decremented depending upon whether the segmentations with which they are consistent are \"popular\" or not.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"These three parsers have given the best reported parsing results on the Penn Treebank Wall Street Journal corpus (Marcus et al., 1993).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Human evaluation is one way to distinguish between the two cases.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We empirically control for the effect of our heuristics to make sure our pruning does not undermine the objectives of our joint task.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"(f1; ;mg n fl1; l2g ; l) 4 (f1; ;m \u00f4\u0080\u0080\u0080 1g n fl1; l2; l3g ; l0) !","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Process statistics are already collected by the kernel (and printing them has no meaningful impact on performance).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Table 9 shows that MADA produces a high quality segmentation, and that the effect of cascading segmentation errors on parsing is only 1.92% F1.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective treebanks is often neglected because of the relative scarcity of problematic constructions.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Finally, Section 5 explains how BABAR relates to previous work, and Section 6 summarizes our conclusions.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Section 3 describes the complete coreference resolution model, which uses the contextual role knowledge as well as more traditional coreference features.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"(2003), which gives 96.8% accuracy on the test set.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We again assume a training set of n examples {x1 . xri} where the first m examples have labels {y1 ... yin}, and the last (n \u2014 m) examples are unlabeled.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"I \u00e2\u0080\u00a2 JAPANS :rl4 .\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\)\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"o\u00c2\u00b7'\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\:J\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7 \u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7'\\; . '.:: ..........0 6.51 9.51 : jj \/ JAPANESE OCTOPUS 10\u00c2\u00b728i\u00c2\u00a3 :_nc HOW SAY f B :rl4 :il: :wen2 t '- \u00e2\u0080\u00a2 :zhang!","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Our system fails in (a) because of$ shenl, a rare family name; the system identifies it as a family name, whereas it should be analyzed as part of the given name.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Our experiments consistently demonstrate that this model architecture yields substantial performance gains over more complex tagging counterparts.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The choice of the genre commentary resulted from the fact that an investigation of rhetorical structure, its interaction with other aspects of discourse structure, and the prospects for its automatic derivation are the key motivations for building up the corpus.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"selected; and that recall is defined to be the number of correct hits divided by the number of items that should have been selected.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We then extract a set of possible tags tx(y) by eliminating labels whose probability is below a threshold value \u03c4: We describe how we choose \u03c4 in \u00a76.4.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"1 74.5 56.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The paper is structured as follows.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"For instance, by altering the emission distribution parameters, Johnson (2007) encourages the model to put most of the probability mass on few tags.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This is in general very difficult, given the extremely free manner in which Chinese given names are formed, and given that in these cases we lack even a family name to give the model confidence that it is identifying a name.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"However, MADA is language-specific and relies on manually constructed dictionaries.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"First we mark any node that dominates (at any level) a verb sider POS tags when pre-terminals are the only intervening nodes between the nucleus and its bracketing (e.g., unaries, base NPs).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"2.2 Contextual Role Knowledge.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This model is easily incorporated into the segmenter by building a WFST restrict\u00c2\u00ad ing the names to the four licit types, with costs on the arcs for any particular name summing to an estimate of the cost of that name.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We refer to (T , W ) as the lexicon of a language and \u00cf\u0088 for the parameters for their generation; \u00cf\u0088 depends on a single hyperparameter \u00ce\u00b2.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"While there are other obstacles to completing this idea, we believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"(Again, the goal of also in structural features.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"4 Evaluation Results.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The similar explanation applies to the link to the \u00e2\u0080\u009cstake\u00e2\u0080\u009d set.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Recently, this topic has been getting more attention, as is evident from the Paraphrase Workshops in 2003 and 2004, driven by the needs of various NLP applications.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We place certain restrictions on the composition operations of LCFRS's, restrictions that are shared by the composition operations of the constrained grammatical systems that we have considered.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In this section, we will explain the algorithm step by step with examples.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"If is not initCaps, then (not-initCaps, ) is set to 1.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Specifically, we (+FEATS) utilizes the tag prior as well as features (e.g., suffixes and orthographic features), discussed in Section 3, for the P (W |T , \u00cf\u0088) component.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Table 2 Similarity matrix for segmentation judgments.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Most languages that use Roman, Greek, Cyrillic, Armenian, or Semitic scripts, and many that use Indian-derived scripts, mark orthographic word boundaries; however, languages written in a Chinese-derived writ\u00c2\u00ad ing system, including Chinese and Japanese, as well as Indian-derived writing systems of languages like Thai, do not delimit orthographic words.1 Put another way, written Chinese simply lacks orthographic words.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In this paper, we will propose an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"It is well-known that constituency parsing models designed for English often do not generalize easily to other languages and treebanks.1 Explanations for this phenomenon have included the relative informativeness of lexicalization (Dubey and Keller, 2003; Arun and Keller, 2005), insensitivity to morphology (Cowan and Collins, 2005; Tsarfaty and Sima\u00e2\u0080\u0099an, 2008), and the effect of variable word order (Collins et al., 1999).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"3.1 Lexicon Component.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Note that in our construction arcs can never cross token boundaries.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Better Arabic Parsing: Baselines, Evaluations, and Analysis","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The simplest version of the maximum matching algorithm effectively deals with ambiguity by ignoring it, since the method is guaranteed to produce only one segmentation.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"However, when we pre- tag the input\u00e2\u0080\u0094as is recommended for English\u00e2\u0080\u0094 we notice a 0.57% F1 improvement.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The restriction can be expressed in terms of the number of uncovered source sentence positions to the left of the rightmost position m in the coverage set.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"If there is a frequent multi-word sequence in a domain, we could use it as a keyword candidate.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"On MUC6 data, MENERGI also achieves performance comparable to IdentiFinder when trained on similar amount of training data.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The key to the methods we describe is redundancy in the unlabeled data.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"For these models we limit the options provided for OOV words by not considering the entire token as a valid segmentation in case at least some prefix segmentation exists.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"3 Techniques for automatic vocalization have been studied (Zitouni et al., 2006; Habash and Rambow, 2007).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The results clearly indicate increased variation in the ATB relative to the WSJ, but care should be taken in assessing the magnitude of the difference.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"DempsterShafer handles this by re-normalizing all the belief values with respect to only the non-null sets (this is the purpose of the denominator in Equation 1).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"We often encounter sentences that are highly ambiguous in themselves, without some prior knowledge of the entities concerned.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"We also see that the increase in the size of the label sets for Head and Head+Path is far below the theoretical upper bounds given in Table 1.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"AdaBoost finds a weighted combination of simple (weak) classifiers, where the weights are chosen to minimize a function that bounds the classification error on a set of training examples.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Mikheev et al.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Central to our approach (see Algorithm 1) is a bilingual similarity graph built from a sentence-aligned parallel corpus.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"gaolxing4 'happy'=> F.i'JF.i'JJI!JI!","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The details are given in (Och and Ney, 2000).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Cur In order to ground such approaches in linguistic observation and description, a multi-level anno 10 For an exposition of the idea as applied to the task of text planning, see (Chiarcos, Stede 2004).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This is in general very difficult, given the extremely free manner in which Chinese given names are formed, and given that in these cases we lack even a family name to give the model confidence that it is identifying a name.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Also, the method of using keywords rules out phrases which don\u00e2\u0080\u0099t contain popular words in the domain.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Sometimes, however, these beliefs can be contradictory.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This information is readily available in TRIE where adjacent records with equal pointers indicate no further extension of context is possible.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Not surprisingly some semantic classes are better for names than others: in our corpora, many names are picked from the GRASS class but very few from the SICKNESS class.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"As discussed in more detail in \u00a73, we use two types of vertices in our graph: on the foreign language side vertices correspond to trigram types, while the vertices on the English side are individual word types.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Our experiments all concern the analytical annotation, and the first experiment is based only on the training part.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"For example, out of 905 phrases in the CC- domain, 211 phrases contain keywords found in step 2.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Now we have sets of phrases which share a keyword and we have links between those sets.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Finally, we would like to note that it is possible to devise similar algorithms based with other objective functions than the one given in Equ.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"ogy (Koskenniemi 1983; Antworth 1990; Tzoukermann and Liberman 1990; Karttunen, Kaplan, and Zaenen 1992; Sproat 1992); we represent the fact that ir, attaches to nouns by allowing t:-transitions from the final states of all noun entries, to the initial state of the sub-WFST representing f,.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"In the rhetorical tree, nuclearity information is then used to extract a \u00e2\u0080\u009ckernel tree\u00e2\u0080\u009d that supposedly represents the key information from which the summary can be generated (which in turn may involve co-reference information, as we want to avoid dangling pronouns in a summary).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"newspaper material, but also including kungfu fiction, Buddhist tracts, and scientific material.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Though we do not directly compare state implementations, performance metrics in Table 1 indicate our overall method is faster.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"All the NE pair instances which co-occur separated by at most 4 chunks are collected along with information about their NE types and the phrase between the NEs (the \u00e2\u0080\u0098context\u00e2\u0080\u0099).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In considering the recognition of these languages, we were forced to be more specific regarding the relationship between the structures derived by these formalisms and the substrings they span.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Thus corresponding pseudo-labels for instances on which gj abstain are set to zero and these instances do not contribute to the objective function.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We can make several observations on the cause of errors.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Smith estimates Lotus will make a profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d, our system extracts \u00e2\u0080\u009cSmith esti mates Lotus\u00e2\u0080\u009d as an instance.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"0 . 8 3 1 0.859 496 76.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"This solution also obviates the need to perform word sense disambiguation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"There is a sizable literature on Chinese word segmentation: recent reviews include Wang, Su, and Mo (1990) and Wu and Tseng (1993).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Reiche\u00e2\u0080\u0099s colleagues will make sure that the concept is waterproof.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Otherwise, label the training data with the combined spelling\/contextual decision list, then induce a final decision list from the labeled examples where all rules (regardless of strength) are added to the decision list.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"This is in contrast to dependency treebanks, e.g.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The algorithm in Fig.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Proper names are assumed to be coreferent if they match exactly, or if they closely match based on a few heuristics.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"As observed by Kahane et al. (1998), any (nonprojective) dependency graph can be transformed into a projective one by a lifting operation, which replaces each non-projective arc wj wk by a projective arc wi \u2014* wk such that wi \u2014*\u2217 wj holds in the original graph.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"By design, they readily capture regularities at the token-level.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Equation 2 is an estimate of the conditional probability of the label given the feature, P(yjx).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"When translating the sentence monotonically from left to right, the translation of the German finite verb 'kann', which is the left verbal brace in this case, is postponed until the German noun phrase 'mein Kollege' is translated, which is the subject of the sentence.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(6), with W+ > W_.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We extend Subramanya et al.\u2019s intuitions to our bilingual setup.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"(2009) study related but different multilingual grammar and tagger induction tasks, where it is assumed that no labeled data at all is available.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For speed, we plan to implement the direct-mapped cache from BerkeleyLM.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Church and Hanks [1989]), and we have used lists of character pairs ranked by mutual information to expand our own dictionary.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Thus the method makes the fairly strong assumption that the features can be partitioned into two types such that each type alone is sufficient for classification.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"This research was supported in part by the Defense Advanced Research Projects Agency as part of the Translingual Information Detection, Extraction and Summarization (TIDES) program, under Grant N66001001-18917 from the Space and Naval Warfare Systems Center, San Diego, and by the National Science Foundation under Grant IIS00325657.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Section 3 discusses the applications that have been completed with PCC, or are under way, or are planned for the future.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"10 Other orthographic normalization schemes have been suggested for Arabic (Habash and Sadat, 2006), but we observe negligible parsing performance differences between these and the simple scheme used in this evaluation.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Put another way, the minimum of Equ.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"2.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Again we notice that the isolated constituent precision is larger than 0.5 only in those partitions that contain very few samples.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Smith estimates Lotus will make profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For a given partial hypothesis (C; j), the order in which the cities in C have been visited can be ignored (except j), only the score for the best path reaching j has to be stored.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The tokens w are generated by token-level tags t from an HMM parameterized by the lexicon structure.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Approaches differ in the algorithms used for scoring and selecting the best path, as well as in the amount of contextual information used in the scoring process.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Also, in Information Extraction (IE), in which the system tries to extract elements of some events (e.g. date and company names of a corporate merger event), several event instances from different news articles have to be aligned even if these are expressed differently.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"This procedure is repeated for T rounds while alternating between the two classifiers.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Words and punctuation that appear in brackets are considered optional.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Finally, we note that Jiang\u2019s instance-weighting framework is broader than we have presented above, encompassing among other possibilities the use of unlabelled IN data, which is applicable to SMT settings where source-only IN corpora are available.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The linear LM (lin lm), TM (lin tm) and MAP TM (map tm) used with non-adapted counterparts perform in all cases slightly worse than the log-linear combination, which adapts both LM and TM components.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"This is similar to stacking the different feature instantiations into long (sparse) vectors and computing the cosine similarity between them.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Sie.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"For the experiments, we use a simple preprocessing step.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The phrases have to be the expressions of length less than 5 chunks, appear between two NEs.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Quite often, though, these directives fulfill the goal of increasing annotator agreement without in fact settling the theoretical question; i.e., the directives are clear but not always very well motivated.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"A name already mentioned previously in a document may appear in abbreviated form when it is mentioned again later.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Also, expanding on the techniques for the automatic generation of extraction patterns (Riloff 96; Sudo 03) using our method, the extraction patterns which have the same meaning can be automatically linked, enabling us to produce the final table fully automatically.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"gaolbu4-gaolxing4 (hap-not-happy) 'happy?'","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The high 1 tone of J1l would not normally neutralize in this fashion if it were functioning as a word on its own.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For each token , zero, one, or more of the features in each feature group are set to 1.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Our coreference resolver performed well in two domains, and experiments showed that each contextual role knowledge source contributed valuable information.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The second row represents the performance of the median hyperparameter setting.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"This is the only important case, because otherwise the simple majority combining technique would pick the correct constituent.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"For example, one parser could be more accurate at predicting noun phrases than the other parsers.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In (Bikel et al., 1997) and (Bikel et al., 1999), performance was plotted against training data size to show how performance improves with training data size.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"2.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"This distributional sparsity of syntactic tags is not unique to English 1 The source code for the work presented in this paper is available at http:\/\/groups.csail.mit.edu\/rbg\/code\/typetagging\/.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"3 61.7 38.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"This measure has the advantage of being completely automatic.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"MENE without Proteus, however, did not do very well and only achieved an F measure of 84.22% (Borthwick, 1999).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"(c) Coordination ambiguity is shown in dependency scores by e.g., \u00e2\u0088\u0097SSS R) and \u00e2\u0088\u0097NP NP NP R).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Ex: He was found in San Jose, where ...","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The kind of annotation work presented here would clearly benefit from the emergence of standard formats and tag sets, which could lead to sharable resources of larger size.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Otherwise for the predecessor search hypothesis, we would have chosen a position that would not have been among the first n uncovered positions.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Gather phrases using keywords Next, we select a keyword for each phrase \u00e2\u0080\u0093 the top-ranked word based on the TF\/IDF metric.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The bias of automatic methods in favor of statistical systems seems to be less pronounced on out-of-domain test data.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Here, the pruning threshold t0 = 10:0 is used.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In a fully supervised setting, the task is to learn a function f such that for all i = 1...m, f (xi,i, 12,i) = yz.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"For example, the two NEs \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d and \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d have the following contexts.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.1.1 Lexical Seeding It is generally not safe to assume that multiple occurrences of a noun phrase refer to the same entity.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In the second and third translation examples, the IbmS word reordering performs worse than the QmS word reordering, since it can not take properly into account the word reordering due to the German verbgroup.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Evaluation of links A link between two sets is considered correct if the majority of phrases in both sets have the same meaning, i.e. if the link indicates paraphrase.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The motivation for our more informal approach was the intuition that there are so many open problems in rhetorical analysis (and more so for German than for English; see below) that the main task is qualitative investigation, whereas rigorous quantitative analyses should be performed at a later stage.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"For RandLM, we used the settings in the documentation: 8 bits per value and false positive probability 1 256.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"And then there are decisions that systems typically hard-wire, because the linguistic motivation for making them is not well understood yet.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"(2009), who also incorporate a sparsity constraint, but does via altering the model objective using posterior regularization.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"At this stage the lattice path corresponds to segments only, with no PoS assigned to them.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We have presented a method for unsupervised part- of-speech tagging that considers a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Another technique for parse hybridization is to use a na\u00efve Bayes classifier to determine which constituents to include in the parse.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Finally, we model the probability of a new transliterated name as the product of PTN and PTN(hanzi;) for each hanzi; in the putative name.13 The foreign name model is implemented as an WFST, which is then summed with the WFST implementing the dictionary, morpho 13 The current model is too simplistic in several respects.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"For Arabic we M o d e l S y s t e m L e n g t h L e a f A n c e s t o r Co rpu s Sent Exact E v a l b L P LR F1 T a g % B a s e l i n e 7 0 St an for d (v 1.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The semantic caseframe expectations are used in two ways.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"1 74.5 56.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The morphological anal\u00c2\u00adysis itself can be handled using well-known techniques from finite-state morphol 9 The initial estimates are derived from the frequencies in the corpus of the strings of hanzi making up.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This procedure is repeated for T rounds while alternating between the two classifiers.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The compact variant uses sorted arrays instead of hash tables within each node, saving some memory, but still stores full 64-bit pointers.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"To define a similarity function between the English and the foreign vertices, we rely on high-confidence word alignments.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"There is a guarantee of no crossing brackets but there is no guarantee that a constituent in the tree has the same children as it had in any of the three original parses.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In addition, the restricted version of CG's (discussed in Section 6) generates tree sets with independent paths and we hope that it can be included in a more general definition of LCFRS's containing formalisms whose tree sets have path sets that are themselves LCFRL's (as in the case of the restricted indexed grammars, and the hierarchy defined by Weir).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We adapt the string pumping lemma for the class of languages corresponding to the complexity of the path set.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Nonstochastic lexical-knowledge-based approaches have been much more numer\u00c2\u00ad ous.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The development of automatic scoring methods is an open field of research.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The maximum entropy framework estimates probabilities based on the principle of making as few assumptions as possible, other than the constraints imposed.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"3 54.4 33.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"This stage of label propagation results in a tag distribution ri over labels y, which encodes the proportion of times the middle word of ui E Vf aligns to English words vy tagged with label y: The second stage consists of running traditional label propagation to propagate labels from these peripheral vertices Vf\ufffd to all foreign language vertices in the graph, optimizing the following objective: 5 POS Induction After running label propagation (LP), we compute tag probabilities for foreign word types x by marginalizing the POS tag distributions of foreign trigrams ui = x\u2212 x x+ over the left and right context words: where the qi (i = 1, ... , |Vf|) are the label distributions over the foreign language vertices and \u00b5 and \u03bd are hyperparameters that we discuss in \u00a76.4.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The probabilities are incorporated into the DempsterShafer model using Equation 1.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Searching a probing hash table consists of hashing the key, indexing the corresponding bucket, and scanning buckets until a matching key is found or an empty bucket is encountered, in which case the key does not exist in the table.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"27 80.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We have already mentioned the closely related work by Matsoukas et al (2009) on discriminative corpus weighting, and Jiang and Zhai (2007) on (nondiscriminative) instance weighting.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thus, the length of any string in L is a linear combination of the length of strings in some fixed finite subset of L, and thus L is said to have the constant growth property.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Equation 2 is an estimate of the conditional probability of the label given the feature, P(yjx).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"yields.6 But to its great advantage, it has a high ratio of non-terminals\/terminals (\u00ce\u00bc Constituents \/ \u00ce\u00bc Length).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Pseudo-code describing the generalized boosting algorithm of Schapire and Singer is given in Figure 1.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Unigrams also have 64-bit overhead for vocabulary lookup.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"In addition to using the simple features directly, we also trained an SVM classifier with these features to distinguish between IN and OUT phrase pairs.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The method halves the error rate in comparison to a method using the labeled examples alone.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Thus, most broad-coverage parsers based on dependency grammar have been restricted to projective structures.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"In graph-based learning approaches one constructs a graph whose vertices are labeled and unlabeled examples, and whose weighted edges encode the degree to which the examples they link have the same label (Zhu et al., 2003).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The question is how to normalize the probabilities in such a way that smaller groupings have a better shot at winning.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"4 To be sure, it is not always true that a hanzi represents a syllable or that it represents a morpheme.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Information from a sentence is sometimes insufficient to classify a name correctly.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Still, for both human and automatic rhetorical analysis, connectives are the most important source of surface information.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We used these three parsers to explore parser combination techniques.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Specifically, for the ith word type, the set of token-level tags associated with token occurrences of this word, denoted t(i), must all take the value Ti to have nonzero mass. Thus in the context of Gibbs sampling, if we want to block sample Ti with t(i), we only need sample values for Ti and consider this setting of t(i).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The semantic agreement KS eliminates some candidates, but also provides positive evidence in one case: if the candidate and anaphor both have semantic tags human, company, date, or location that were assigned via NER or the manually labeled dictionary entries.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Note that we check for , the word preceding the consecutive sequence of initCaps tokens, since person prefixes like Mr., Dr., etc are not part of person names, whereas corporate suffixes like Corp., Inc., etc are part of corporate names.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"By sharing stacks (in IG's) or by using nonlinear equations over f-structures (in FUG's and LFG's), structures with unbounded dependencies between paths can be generated.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"For each domain, phrases which contain the same keyword are gathered to build a set of phrases (Step 3).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We of course also fail to identify, by the methods just described, given names used without their associated family name.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"For Arabic we M o d e l S y s t e m L e n g t h L e a f A n c e s t o r Co rpu s Sent Exact E v a l b L P LR F1 T a g % B a s e l i n e 7 0 St an for d (v 1.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This is similar to using the Linux MAP POPULATE flag that is our default loading mechanism.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"And time is short.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Thus corresponding pseudo-labels for instances on which gj abstain are set to zero and these instances do not contribute to the objective function.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The dev and test sets were randomly chosen from the EMEA corpus.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Dictionaries: Due to the limited amount of training material, name dictionaries have been found to be useful in the named entity task.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"This work was funded by NSF grant IRI-9502312.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"There has also been work using a bootstrap- ping approach [Brin 98; Agichtein and Gravano 00; Ravichandran and Hovy 02].","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Otherwise for the predecessor search hypothesis, we would have chosen a position that would not have been among the first n uncovered positions.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Different sentence structure and rich target language morphology are two reasons for this.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The kind of annotation work presented here would clearly benefit from the emergence of standard formats and tag sets, which could lead to sharable resources of larger size.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"DempsterShafer handles this by re-normalizing all the belief values with respect to only the non-null sets (this is the purpose of the denominator in Equation 1).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Top 5 Bot to m 5 Go ld NN P NN JJ CD NN S RB S PD T # \u00e2\u0080\u009d , 1T W CD W RB NN S VB N NN PR P$ W DT : MD . +P RI OR CD JJ NN S WP $ NN RR B- , $ \u00e2\u0080\u009d . +F EA TS JJ NN S CD NN P UH , PR P$ # . \u00e2\u0080\u009c Table 5: Type-level English POS Tag Ranking: We list the top 5 and bottom 5 POS tags in the lexicon and the predictions of our models under the best hyperparameter setting.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"An examination of the subjects' bracketings confirmed that these instructions were satisfactory in yielding plausible word-sized units.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The remaining arcs are marked OOV.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(1992).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The method reported in this paper makes use solely of unigram probabilities, and is therefore a zeroeth-order model: the cost of a particular segmentation is estimated as the sum of the costs of the individual words in the segmentation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"att.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Our first model is GTplain, a PCFG learned from the treebank after removing all functional features from the syntactic categories.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Experiments using data from the Prague Dependency Treebank show that the combined system can handle nonprojective constructions with a precision sufficient to yield a significant improvement in overall parsing accuracy.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"On the other hand, we can expect Head+Path to be the most useful representation for reconstructing the underlying non-projective dependency graph.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Similar behavior is observed when adding features.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"0 55.3 34.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The dev and test sets were randomly chosen from the EMEA corpus.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"In focus is in particular the correlation with rhetorical structure, i.e., the question whether specific rhetorical relations \u00e2\u0080\u0094 or groups of relations in particular configurations \u00e2\u0080\u0094 are signalled by speakers with prosodic means.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Table 4 Differences in performance between our system and Wang, Li, and Chang (1992).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Label propagation can then be used to transfer the labels to the peripheral foreign vertices (i.e. the ones adjacent to the English vertices) first, and then among all of the foreign vertices (\u00a74).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"For each feature type f and tag t, a multinomial \u00cf\u0088tf is drawn from a symmetric Dirichlet distribution with concentration parameter \u00ce\u00b2.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Therefore we opted to use the much 7 The DempsterShafer theory assumes that one of the hypotheses in \u00ce\u00b8 is correct, so eliminating all of the hypotheses violates this assumption.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We have not to date explored these various options.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We do not experiment with models larger than physical memory in this paper because TPT is unreleased, factors such as disk speed are hard to replicate, and in such situations we recommend switching to a more compact representation, such as RandLM.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"It is a relatively frequent word in the domain, but it can be used in different extraction scenarios.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"One conclusion drawn from this annotation effort was that for humans and machines alike, 2 www.sfs.nphil.unituebingen.de\/Elwis\/stts\/ stts.html 3 www.coli.unisb.de\/sfb378\/negra-corpus\/annotate.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"0 . 8 3 1 0.859 496 76.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In the named entity domain these rules were Each of these rules was given a strength of 0.9999.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"1 61.2 43.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, there is a strong relationship between ni1s and the number of hanzi in the class.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"To differentiate between the coordinating and discourse separator functions of conjunctions (Table 3), we mark each CC with the label of its right sister (splitCC).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"2.2.3 Lexical Caseframe Expectations The second type of contextual role knowledge learned by BABAR is Lexical Caseframe Expectations, which are used by the CFLex knowledge source.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"i=1 (f,v)\u00e2\u0088\u0088Wi","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"1,000 of these were picked at random, and labeled by hand to produce a test set.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The alignment model uses two kinds of parameters: alignment probabilities p(aj jaj\u00f4\u0080\u0080\u00801; I; J), where the probability of alignment aj for position j depends on the previous alignment position aj\u00f4\u0080\u0080\u00801 (Ney et al., 2000) and lexicon probabilities p(fj jeaj ).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Contains digits.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Without using the same test corpus, direct comparison is obviously difficult; fortunately, Chang et al. include a list of about 60 sentence fragments that exemplify various categories of performance for their system.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Our results show that MENERGI performs as well as IdentiFinder when trained on comparable amount of training data.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"To check for this, we do pairwise bootstrap resampling: Again, we repeatedly sample sets of sentences, this time from both systems, and compare their BLEU scores on these sets.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Because conjunctions are elevated in the parse trees when they separate recursive constituents, we choose the right sister instead of the category of the next word.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We briefly review it here for completeness.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We picked two domains, the CC-domain and the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain (PC-domain), for the evaluation, as the entire system output was too large to evaluate.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"( b ) s u p p o r t s c a n d i d a t e i f s e l e c t e d s e m a n t i c t a g s m a t c h t h o s e o f t h e a n a p h o r . Le xic al computes degree of lexical overlap b e t w e e n t h e c a n d i d a t e a n d t h e a n a p h o r . Re cen cy computes the relative distance between the c a n d i d a t e a n d t h e a n a p h o r . Sy nR ole computes relative frequency with which the c a n d i d a t e \u00e2\u0080\u0099 s s y n t a c t i c r o l e o c c u r s i n r e s o l u t i o n s . Figure 4: General Knowledge Sources The Lexical KS returns 1 if the candidate and anaphor are identical, 0.5 if their head nouns match, and 0 otherwise.","label":0},{"source_text":"There is no global pruning.","target_text":"Search hypotheses are processed separately according to their coverage set C. The best scored hypothesis for each coverage set is computed: QBeam(C) = max e;e0 ;S;j Qe0 (e; S; C; j) The hypothesis (e0; e; S; C; j) is pruned if: Qe0 (e; S; C; j) < t0 QBeam(C); where t0 is a threshold to control the number of surviving hypotheses.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The web-based Annis imports data in a variety of XML formats and tagsets and displays it in a tier-orientedway (optionally, trees can be drawn more ele gantly in a separate window).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"If the key distribution\u2019s range is also known (i.e. vocabulary identifiers range from 0 to the number of words), then interpolation search can use this information instead of reading A[0] and A[|A |\u2212 1] to estimate pivots; this optimization alone led to a 24% speed improvement.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"As each global feature group is added to the list of features, we see improvements to both MUC6 and MUC6 MUC7 Baseline 90.75% 85.22% + ICOC 91.50% 86.24% + CSPP 92.89% 86.96% + ACRO 93.04% 86.99% + SOIC 93.25% 87.22% + UNIQ 93.27% 87.24% Table 3: F-measure after successive addition of each global feature group Table 5: Comparison of results for MUC6 Systems MUC6 MUC7 No.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"When the parser is trained on the transformed data, it will ideally learn not only to construct projective dependency structures but also to assign arc labels that encode information about lifts.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Not all the layers have been produced for all the texts yet.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Although this feature helps, we encounter one consequence of variable word order.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"mark- ContainsVerb is especially effective for distinguishing root S nodes of equational sentences.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"MADA uses an ensemble of SVMs to first re-rank the output of a deterministic morphological analyzer.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"A straightforward way to find the shortest tour is by trying all possible permutations of the n cities.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The interdependence between fb or 1\/!i, and 5:2 is not captured by our model, but this could easily be remedied.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Some relations are signalled by subordinating conjunctions, which clearly demarcate the range of the text spans related (matrix clause, embedded clause).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(Hearst 92) describes a method for extracting hyponyms from a corpus (pairs of words in &quot;isa&quot; relations).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"8 1 8.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"For the translation experiments, Eq. 2 is recursively evaluated.","label":0},{"source_text":"Here we present two algorithms.","target_text":"By this assumption, each element x E X can also be represented as (xi, x2) E X1 x X2.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Clearly this is not the only way to estimate word-frequencies, however, and one could consider applying other methods: in partic\u00c2\u00ad ular since the problem is similar to the problem of assigning part-of-speech tags to an untagged corpus given a lexicon and some initial estimate of the a priori probabilities for the tags, one might consider a more sophisticated approach such as that described in Kupiec (1992); one could also use methods that depend on a small hand-tagged seed corpus, as suggested by one reviewer.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Finally, quite a few hanzi are homographs, meaning that they may be pronounced in several different ways, and in extreme cases apparently represent different morphemes: The prenominal modifi\u00c2\u00ad cation marker eg deO is presumably a different morpheme from the second morpheme of \u00c2\u00a7eg mu4di4, even though they are written the same way.4 The second point, which will be relevant in the discussion of personal names in Section 4.4, relates to the internal structure of hanzi.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Step 3.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"This technique has the advantage of requiring no training, but it has the disadvantage of treating all parsers equally even though they may have differing accuracies or may specialize in modeling different phenomena.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This WFST is then summed with the WFST implementing the dictionary and morphological rules, and the transitive closure of the resulting transducer is computed; see Pereira, Riley, and Sproat (1994) for an explanation of the notion of summing WFSTs.12 Conceptual Improvements over Chang et al.'s Model.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"M works in space S(n) if for every string that M accepts no configuration exceeds space S(n).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"However, TRIE partitions storage by n-gram length, so walking the trie reads N disjoint pages.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The Grammar Our parser looks for the most likely tree spanning a single path through the lattice of which the yield is a sequence of lexemes.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"This technique has the advantage of requiring no training, but it has the disadvantage of treating all parsers equally even though they may have differing accuracies or may specialize in modeling different phenomena.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"None of the models attach the attributive adjectives correctly.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Also, \u00e2\u0080\u009cagree\u00e2\u0080\u009d in the CC-domain is not a desirable keyword.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In this paper, we have presented a new, e\u00c3\u0086cient DP-based search procedure for statistical machine translation.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The first two rows of the table are baselines.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Find keywords for each NE pair When we look at the contexts for each domain, we noticed that there is one or a few important words which indicate the relation between the NEs (for example, the word \u00e2\u0080\u009cunit\u00e2\u0080\u009d for the phrase \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":".","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"In a model we built with default settings, 1.2% of n + 1-grams were missing their ngram suffix.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Aside from adding a simple rule to correct alif deletion caused by the preposition J, no other language-specific processing is performed.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The effect of UNIQ is very small on both data sets.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Thus our proposed model is a proper model assigning probability mass to all (7r, L) pairs, where 7r is a parse tree and L is the one and only lattice that a sequence of characters (and spaces) W over our alpha-beth gives rise to.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"A more forceful approach for encoding sparsity is posterior regularization, which constrains the posterior to have a small number of expected tag assignments (Grac\u00c2\u00b8a et al., 2009).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In our second model GTvpi we also distinguished finite and non-finite verbs and VPs as 10Lattice parsing can be performed by special initialization of the chart in a CKY parser (Chappelier et al., 1999).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Table 6: Example Translations for the Verbmobil task.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In Eq.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Central to our approach (see Algorithm 1) is a bilingual similarity graph built from a sentence-aligned parallel corpus.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"9 www.ling.unipotsdam.de\/sfb\/ Figure 2: Screenshot of Annis Linguistic Database 3.3 Symbolic and knowledge-based.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"These systems are similar to those described by Pollard (1984) as Generalized Context-Free Grammars (GCFG's).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"08 84.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"With each iteration more examples are assigned labels by both classifiers, while a high level of agreement (> 94%) is maintained between them.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Both parameters depend on a single hyperparameter \u00ce\u00b1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Since we could not bias the subjects towards a particular segmentation and did not presume linguistic sophistication on their part, the instructions were simple: subjects were to mark all places they might plausibly pause if they were reading the text aloud.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"SRILM (Stolcke, 2002) is widely used within academia.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"4 70.4 46.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The approach builds from an initial seed set for a category, and is quite similar to the decision list approach described in (Yarowsky 95).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2010) consistently outperforms ours on English, we obtain substantial gains across other languages.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"While our method also enforces a singe tag per word constraint, it leverages the transition distribution encoded in an HMM, thereby benefiting from a richer representation of context.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"TRIE uses less memory and has better locality.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"For all grammars, we use fine-grained PoS tags indicating various morphological features annotated therein.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"First, the model assumes independence between the first and second hanzi of a double given name.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Alternatively, h can be thought of as defining a decision list of rules x y ranked by their &quot;strength&quot; h(x, y).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Table 2 Similarity matrix for segmentation judgments.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"(Levinger et al., 1995; Goldberg et al., ; Adler et al., 2008)) will make the parser more robust and suitable for use in more realistic scenarios.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The translation direction is from German to English.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The DempsterShafer rule for combining pdfs is: to {C}, meaning that it is 70% sure the correct hypothesis is C. The intersection of these sets is the null set because these beliefs are contradictory.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In total 13,976 phrases are assigned to sets of phrases, and the accuracy on our evaluation data ranges from 65 to 99%, depending on the domain and the size of the sets.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"including Third Tone Sandhi (Shih 1986), which changes a 3 (low) tone into a 2 (rising) tone before another 3 tone: 'j\";gil, xiao3 [lao3 shu3] 'little rat,' becomes xiao3 { lao2shu3 ], rather than xiao2 { lao2shu3 ], because the rule first applies within the word lao3shu3 'rat,' blocking its phrasal application.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":1},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Caching for IRSTLM is smaller at 0.09 GB resident memory, though it supports only a single thread.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"However, we have reason to doubt Chang et al.'s performance claims.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"pronunciation depends upon word affiliation: tfJ is pronounced deO when it is a prenominal modification marker, but di4 in the word \u00c2\u00a7tfJ mu4di4 'goal'; fl; is normally ganl 'dry,' but qian2 in a person's given name.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thus, any language that is letter equivalent to a semilinear language is also semilinear.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"123 examples fell into the noise category.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Before we turn to the evaluation, however, we need to introduce the data-driven dependency parser used in the latter experiments.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Note that in our model, conditioned on T , there is precisely one t which has nonzero probability for the token component, since for each word, exactly one \u00ce\u00b8t has support.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The additional morphological material in such cases appears after the stem and realizes the extended meaning.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Hence, we take the probability of the event fmnh analyzed as REL VB to be This means that we generate f and mnh independently depending on their corresponding PoS tags, and the context (as well as the syntactic relation between the two) is modeled via the derivation resulting in a sequence REL VB spanning the form fmnh. based on linear context.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In addition to the automatic methods, AG, GR, and ST, just discussed, we also added to the plot the values for the current algorithm using only dictionary entries (i.e., no productively derived words or names).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"A number of PCC commentaries will be read by professional news speakers and prosodic features be annotated, so that the various annotation layers can be set into correspondence with intonation patterns.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"We call this pseudoprojective dependency parsing, since it is based on a notion of pseudo-projectivity (Kahane et al., 1998).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Therefore in cases where the segmentation is identical between the two systems we assume that tagging is also identical.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"They return a value in the range [0,1], where 0 indicates neutrality and 1 indicates the strongest belief that the candidate and anaphor are coreferent.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"rently, some annotations (in particular the connectives and scopes) have already moved beyond the core corpus; the others will grow step by step.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"4.1 Dictionary Representation.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The general idea for the knowledge- based part is to have the system use as much information as it can find at its disposal to produce a target representation as specific as possible and as underspecified as necessary.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"CFG's, TAG's, MCTAG's and HG's are all members of this class since they satisfy these restrictions.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"MENE without Proteus, however, did not do very well and only achieved an F measure of 84.22% (Borthwick, 1999).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Translation errors are reported in terms of multireference word error rate (mWER) and subjective sentence error rate (SSER).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Table 3 Classes of words found by ST for the test corpus.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"76 16.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"76 16.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"However, the data sparsity induced by vocalization makes it difficult to train statistical models on corpora of the size of the ATB, so vocalizing and then parsing may well not help performance.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Initially, we planned to compare the semantic classes of an anaphor and a candidate and infer that they might be coreferent if their semantic classes intersected.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"42 nator, the N31s can be measured well by counting, and we replace the expectation by the observation.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"IL+-1Proof: Assume a pair of crossing constituents appears in the output of the constituent voting technique using k parsers.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Secondly, for all our models we provide better fine- and coarse-grained POS-tagging accuracy, and all pruned models outperform the Oracle results reported by them.12 In terms of syntactic disambiguation, even the simplest grammar pruned with HSPELL outperforms their non-Oracle results.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"If they knew that the first four words in a hypergraph node would never extend to the left and form a 5-gram, then three or even fewer words could be kept in the backward state.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"If one is interested in translation, one would probably want to consider show up as a single dictionary word since its semantic interpretation is not trivially derivable from the meanings of show and up.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"pre-processing.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Making the reasonable assumption that similar information is relevant for solving these problems in Chinese, it follows that a prerequisite for intonation-boundary assignment and prominence assignment is word segmentation.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"yu2 'fish.'","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"An inverted alignment is defined as follows: inverted alignment: i ! j = bi: Target positions i are mapped to source positions bi.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Our approach uses word-to-word dependencies between source and target words.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Before describing the unsupervised case we first describe the supervised version of the algorithm: Input to the learning algorithm: n labeled examples of the form (xi, y\u201e). y, is the label of the ith example (given that there are k possible labels, y, is a member of y = {1 ... 0). xi is a set of mi features {x,1, Xi2 .","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Each of the constituents must have received at least 1 votes from the k parsers, so a > I1 and 2 \u2014 2k\u00b1-1 b > ri-5-111.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We evaluated BABAR on two domains: terrorism and natural disasters.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In our model, we associate these features at the type-level in the lexicon.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"As indicated by bolding, for seven out of eight languages the improvements of the \u201cWith LP\u201d setting are statistically significant with respect to the other models, including the \u201cNo LP\u201d setting.11 Overall, it performs 10.4% better than the hitherto state-of-the-art feature-HMM baseline, and 4.6% better than direct projection, when we macro-average the accuracy over all languages.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Before presenting our results, we describe the datasets that we used, as well as two baselines.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For terrorism, BABAR generated 5,078 resolutions: 2,386 from lexical seeding and 2,692 from syntactic seeding.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The first four affixes are so-called resultative affixes: they denote some prop\u00c2\u00ad erty of the resultant state of a verb, as in E7 wang4bu4-liao3 (forget-not-attain) 'cannot forget.'","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This representation gives ir, an appropriate morphological decomposition, pre\u00c2\u00ad serving information that would be lost by simply listing ir, as an unanalyzed form.","label":0},{"source_text":"There is no global pruning.","target_text":"The effect of the pruning threshold t0 is shown in Table 5.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"-1 means that an NP should be ruled out as a possible antecedent, and 0 means that the knowledge source remains neutral (i.e., it has no reason to believe that they cannot be coreferent).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The task can be considered to be one component of the MUC (MUC-6, 1995) named entity task (the other task is that of segmentation, i.e., pulling possible people, places and locations from text before sending them to the classifier).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"But in most cases they can be used interchangably.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"8 66.4 52.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We call this technique constituent voting.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The hypothesized parse is then the set of constituents that are likely (P > 0.5) to be in the parse according to this model.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Instead of offsetting new topics with punctuation, writers of MSA in sert connectives such as \u00ef\u00bf\u00bd wa and \u00ef\u00bf\u00bd fa to link new elements to both preceding clauses and the text as a whole.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"and \u00e2\u0080\u009cH\u00e2\u0080\u009d represents \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"All our results are obtained by using only the official training data provided by the MUC conferences.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For sorted lookup, we compare interpolation search, standard C++ binary search, and standard C++ set based on red-black trees.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In Eq.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"3 54.4 33.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In order to create an IE system for a new domain, one has to spend a long time to create the knowledge.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"[Hasegawa et al. 04] reported only on relation discovery, but one could easily acquire para phrases from the results.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"gaolxing4 'happy'=> F.i'JF.i'JJI!JI!","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In the denomi 11 We have two such lists, one containing about 17,000 full names, and another containing frequencies of.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In a few cases, the criteria for correctness are made more explicit.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Input: Ja , wunderbar . Konnen wir machen . MonS: Yes, wonderful.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The first four affixes are so-called resultative affixes: they denote some prop\u00c2\u00ad erty of the resultant state of a verb, as in E7 wang4bu4-liao3 (forget-not-attain) 'cannot forget.'","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"For example, in the CC-domain, 96 keywords are found which have TF\/ITF scores above a threshold; some of them are shown in Figure 3.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Morphological processes in Semitic languages deliver space-delimited words which introduce multiple, distinct, syntactic units into the structure of the input sentence.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"na me =>2 ha nzi fa mi ly 1 ha nzi gi ve n 6.1 ha nzi fa mi ly => ha nz ii 7.2 ha nzi fa mi ly => ha nzi i ha nz ij 8.1 ha nzi gi ve n => ha nz ii 9.2 ha nzi giv en => ha nzi i ha nz ij The difficulty is that given names can consist, in principle, of any hanzi or pair of hanzi, so the possible given names are limited only by the total number of hanzi, though some hanzi are certainly far more likely than others.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(4) gives In order to minimize Zt, at each iteration the final algorithm should choose the weak hypothesis (i.e., a feature xt) which has values for W+ and W_ that minimize Equ.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"However, there is a crucial difference: the morphological probabilities in their model come from discriminative models based on linear context.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"60 ( P e tr o v, 2 0 0 9 ) all B e r k e l e y ( S e p . 0 9 ) B a s e l i n e 7 0 a l l G o l d P O S 70 \u00e2\u0080\u0094 \u00e2\u0080\u0094 \u00e2\u0080\u0094 0 . 8 0 9 0.839 335 0 . 7 9","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"To date we have not done a separate evaluation of foreign-name recognition.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We also mark all nodes that dominate an SVO configuration (containsSVO).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Each of the constituents must have received at least 1 votes from the k parsers, so a > I1 and 2 \u2014 2k\u00b1-1 b > ri-5-111.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The second stage links sets which involve the same pairs of individual NEs.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"For each pair we also record the context, i.e. the phrase between the two NEs (Step1).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Each feature group can be made up of many binary features.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"10 Here we use the Good-Turing estimate (Baayen 1989; Church and Gale 1991), whereby the aggregate probability of previously unseen instances of a construction is estimated as ni\/N, where N is the total number of observed tokens and n1 is the number of types observed only once.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We also removed the context-type feature type when using the CoBoost approach.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Label propagation is used to propagate these tags inwards and results in tag distributions for the middle word of each Italian trigram.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Simply assigning to each word its most frequent associated tag in a corpus achieves 94.6% accuracy on the WSJ portion of the Penn Treebank.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We then use linguistic and annotation insights to develop a manually annotated grammar for Arabic (\u00c2\u00a74).","label":1},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The features are used to represent each example for the learning algorithm.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"None of the models attach the attributive adjectives correctly.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The use of weighted transducers in particular has the attractive property that the model, as it stands, can be straightforwardly interfaced to other modules of a larger speech or natural language system: presumably one does not want to segment Chinese text for its own sake but instead with a larger purpose in mind.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The corpus was wordaligned using both HMM and IBM2 models, and the phrase table was the union of phrases extracted from these separate alignments, with a length limit of 7.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"It falls short of the \u201cProjection\u201d baseline for German, but is statistically indistinguishable in terms of accuracy.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"4 53.7 43.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"In both cases the investigators were able to achieve significant improvements over the previous best tagging results.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"This representation of structures by substrings and the composition operation by its effect on substrings is related to the work of Rounds (1985).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Figure 1: Translation of PCC sample commentary (STTS)2.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In fact, during the first rounds many of the predictions of Th., g2 are zero.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Sometimes extracted phrases by themselves are not meaningful to consider without context, but we set the following criteria.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Storing state therefore becomes a time-space tradeoff; for example, we store state with partial hypotheses in Moses but not with each phrase.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The correct resolution in sentence (c) depends on knowledge that kidnappers frequently blindfold their victims.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We have argued that the proposed method performs well.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The original tag set for the CoNLL-X Dutch data set consists of compounded tags that are used to tag multi-word units (MWUs) resulting in a tag set of over 300 tags.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, kidnapping victims should be extracted from the subject of the verb \u00e2\u0080\u009ckidnapped\u00e2\u0080\u009d when it occurs in the passive voice (the shorthand representation of this pattern would be \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We did have a problem classifying some modified noun phrases where the modified phrase does not represent a qualified or restricted form of the head, like \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, as these are both represented by the keyword \u00e2\u0080\u009cchairman\u00e2\u0080\u009d.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This has led previous workers to adopt ad hoc linear weighting schemes (Finch and Sumita, 2008; Foster and Kuhn, 2007; L\u00a8u et al., 2007).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"If two phrases can be used to express the same relationship within an information extraction application (\u00e2\u0080\u009cscenario\u00e2\u0080\u009d), these two phrases are paraphrases.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"so that 'door' would be and in this case the hanzi 7C, does not represent a syllable.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"An error count of 0:0 is assigned to a perfect translation, and an error count of 1:0 is assigned to a semantically and syntactically wrong translation.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The developer explained that the loading process requires extra memory that it then frees. eBased on the ratio to SRI\u2019s speed reported in Guthrie and Hepple (2010) under different conditions.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The perplexity for the trigram language model used is 26:5.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The graph was constructed using 2 million trigrams; we chose these by truncating the parallel datasets up to the number of sentence pairs that contained 2 million trigrams.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For pronouns, however, all of the knowledge sources increased recall, often substantially, and with little if any decrease in precision.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Another attempt at using global information can be found in (Borthwick, 1999).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"These results are achieved by training on the official MUC6 and MUC7 training data, which is much less training data than is used by other machine learning systems that worked on the MUC6 or MUC7 named entity task (Bikel et al., 1997; Bikel et al., 1999; Borth- wick, 1999).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Here, \u00e2\u0080\u009cEG\u00e2\u0080\u009d represents \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"1 2 3.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"PoS tags impose a unique morphological segmentation on surface tokens and present a unique valid yield for syntactic trees.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"4 70.4 46.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"(f1; ;mg n fl1; l2g ; l) 4 (f1; ;m \u00f4\u0080\u0080\u0080 1g n fl1; l2; l3g ; l0) !","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"There has also been work using a bootstrap- ping approach [Brin 98; Agichtein and Gravano 00; Ravichandran and Hovy 02].","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Any opinions, findings, conclusions, or recommendations expressed in this paper are those of the authors, and do not necessarily reflect the views of the funding organizations.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Values in the trie are minimally sized at the bit level, improving memory consumption over trie implementations in SRILM, IRSTLM, and BerkeleyLM.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Assuming unseen objects within each class are equiprobable, their probabilities are given by the Good-Turing theorem as: cis E( n'J.ls) Po oc N * E(N8ls) (2) where p815 is the probability of one unseen hanzi in class cls, E(n'J.15 ) is the expected number of hanzi in cls seen once, N is the total number of hanzi, and E(N(\/ 5 ) is the expected number of unseen hanzi in class cls.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Two issues distinguish the various proposals.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We are not claiming that this method is almighty.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"By establishing significantly higher parsing baselines, we have shown that Arabic parsing performance is not as poor as previously thought, but remains much lower than English.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"We developed a first version of annotation guidelines for co-reference in PCC (Gross 2003), which served as basis for annotating the core corpus but have not been empirically evaluated for inter-annotator agreement yet.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"35 76.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We develop our POS induction model based on the feature-based HMM of Berg-Kirkpatrick et al. (2010).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"A number of PCC commentaries will be read by professional news speakers and prosodic features be annotated, so that the various annotation layers can be set into correspondence with intonation patterns.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Sorted arrays store key-value pairs in an array sorted by key, incurring no space overhead.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(If fewer than n rules have Precision greater than pin, we 3Note that taking tlie top n most frequent rules already makes the method robut to low count events, hence we do not use smoothing, allowing low-count high-precision features to be chosen on later iterations. keep only those rules which exceed the precision threshold.) pm,n was fixed at 0.95 in all experiments in this paper.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"MENE (Maximum Entropy Named Entity) (Borth- wick, 1999) was combined with Proteus (a hand- coded system), and came in fourth among all MUC 7 participants.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"2.5 Connectives with scopes.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We see from these results that the behavior of the parametric techniques are robust in the presence of a poor parser.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"So, there is a limitation that IE can only be performed for a predefined task, like \u00e2\u0080\u009ccorporate mergers\u00e2\u0080\u009d or \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Similarly, the tokens and are tested against each list, and if found, a corresponding feature will be set to 1.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Here, $ is the sentence boundary symbol, which is thought to be at position 0 in the target sentence.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Both parameters depend on a single hyperparameter \u00ce\u00b1.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For unaligned words, we set the tag to the most frequent tag in the corresponding treebank.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Our empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"When a collision occurs, linear probing places the entry to be inserted in the next (higher index) empty bucket, wrapping around as necessary.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The NE tagger is a rule-based system with 140 NE categories [Sekine et al. 2004].","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Again, the idea is that having a picture of syntax, co-reference, and sentence-internal information structure at one\u00e2\u0080\u0099s disposal should aid in finding models of discourse structure that are more explanatory and can be empirically supported.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We received submissions from 14 groups from 11 institutions, as listed in Figure 2.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"2 62.2 39.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"\u00e2\u0088\u0097NP NP PP R) and \u00e2\u0088\u0097NP NP ADJP R) are both iDafa attachment.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Gan's solution depends upon a fairly sophisticated language model that attempts to find valid syntactic, semantic, and lexical relations between objects of various linguistic types (hanzi, words, phrases).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In speech recognition the arcs of the lattice are typically weighted in order to indicate the probability of specific transitions.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"73 81.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"It was then tested on section 22 of the Treebank in conjunction with the other parsers.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"To date we have not done a separate evaluation of foreign-name recognition.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Interestingly, Chang et al. report 80.67% recall and 91.87% precision on an 11,000 word corpus: seemingly, our system finds as many names as their system, but with four times as many false hits.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We tokenize MWUs and their POS tags; this reduces the tag set size to 12.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Email: cls@bell-labs.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Papers that use this method or minor variants thereof include Liang (1986), Li et al.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Hence, the terminal symbols appearing in the structures that are composed are not lost (though a constant number of new symbols may be introduced).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"A position is presented by the word at that position.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Lossy compressed models RandLM (Talbot and Osborne, 2007) and Sheffield (Guthrie and Hepple, 2010) offer better memory consumption at the expense of CPU and accuracy.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Table 3: Training and test conditions for the Verbmobil task (*number of words without punctuation marks).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"BABAR uses a DempsterShafer decision model (Stefik, 1995) to combine the evidence provided by the knowledge sources.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"AdaBoost.MH can be applied to the problem using these pseudolabels in place of supervised examples.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"4.1 Local Features.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For a given \"word\" in the automatic segmentation, if at least k of the hu\u00c2\u00ad man judges agree that this is a word, then that word is considered to be correct.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Statistics for all data sets are shown in Table 2.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Limitations There are several limitations in the methods.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"9 61.0 44.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"When finished, the whole material is written into an XML-structured annotation file.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"x EG, has agreed to be bought by H x EG, now owned by H x H to acquire EG x H\u00e2\u0080\u0099s agreement to buy EG Three of those phrases are actually paraphrases, but sometime there could be some noise; such as the second phrase above.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We excluded these from the evaluation as they can be easily identified with a list of days\/months.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The type of alignment we have considered so far requires the same length for source and target sentence, i.e. I = J. Evidently, this is an unrealistic assumption, therefore we extend the concept of inverted alignments as follows: When adding a new position to the coverage set C, we might generate either \u00c3\u0086 = 0 or \u00c3\u0086 = 1 new target words.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"However, in order to capture the properties of various grammatical systems under consideration, our notation is more restrictive that ILFP, which was designed as a general logical notation to characterize the complete class of languages that are recognizable in polynomial time.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"In each figure the upper graph shows the isolated constituent precision and the bottom graph shows the corresponding number of hypothesized constituents.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We also performed experiments to evaluate the impact of each type of contextual role knowledge separately.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If all three sentences are in the same document, then even a human will find it difficult to classify McCann in (1) into either person or organization, unless there is some other information provided.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"To quantize, we use the binning method (Federico and Bertoldi, 2006) that sorts values, divides into equally sized bins, and averages within each bin.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The relativizer f(\u201cthat\u201d) for example, may attach to an arbitrarily long relative clause that goes beyond token boundaries.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The probability of the classes assigned to the words in a sentence in a document is defined as follows: where is determined by the maximum entropy classifier.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Keyword detection error Even if a keyword consists of a single word, there are words which are not desirable as keywords for a domain.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"By applying an inverse transformation to the output of the parser, arcs with non-standard labels can be lowered to their proper place in the dependency graph, giving rise 1The dependency graph has been modified to make the final period a dependent of the main verb instead of being a dependent of a special root node for the sentence. to non-projective structures.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The resulting parsers surpass the best previously published performance results for the Penn Treebank.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"These are shown, with their associated costs, as follows: ABj nc 4.0 AB C\/jj 6.0 CD \/vb 5.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"LCFRS's have only been loosely defined in this paper; we have yet to provide a complete set of formal properties associated with members of this class.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Each set is assigned two values: belief and plausibility.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Step 3.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"3.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Given the bilingual graph described in the previous section, we can use label propagation to project the English POS labels to the foreign language.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"A very small excerpt from an Italian-English graph is shown in Figure 1.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The 1st block contains the simple baselines from section 2.1.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Consider the case where IX].","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Vocabulary lookup is a hash table mapping from word to vocabulary index.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"It reads a file with a list of German connectives, and when a text is opened for annotation, it highlights all the words that show up in this list; these will be all the potential connectives.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"There are simply not enough votes remaining to allow any of the crossing structures to enter the hypothesized constituent set.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Reference resolution involves finding words that co-refer to the same entity.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Now, for this application one might be tempted to simply bypass the segmentation problem and pronounce the text character-by-character.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"It was then tested on section 22 of the Treebank in conjunction with the other parsers.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The dev and test sets were randomly chosen from the EMEA corpus.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"This section describes an algorithm based on boosting algorithms, which were previously developed for supervised machine learning problems.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(See also Wu and Fung [1994].)","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"As observed by Kahane et al. (1998), any (nonprojective) dependency graph can be transformed into a projective one by a lifting operation, which replaces each non-projective arc wj wk by a projective arc wi \u2014* wk such that wi \u2014*\u2217 wj holds in the original graph.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Table 1 shows four words \u00e2\u0080\u009c 0 Indeed NN Indeed Saddamwhose unvocalized surface forms 0 an are indistinguishable.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We substantially outperform all of them on query speed and offer lower memory consumption than lossless alternatives.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"State will ultimately be used as context in a subsequent query.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Annotation of syntactic structure for the core corpus has just begun.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, both a chair and a truck would be labeled as artifacts, but this does not at all suggest that they are coreferent.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In the first part of the experiment, dependency graphs from the treebanks were projectivized using the algorithm described in section 2.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"As we have said, parse quality decreases with sentence length.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We also removed the context-type feature type when using the CoBoost approach.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"2 61.7 64.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Since Dane\u00cb\u0087s\u00e2\u0080\u0099 proposals of \u00e2\u0080\u0098thematic development patterns\u00e2\u0080\u0099, a few suggestions have been made as to the existence of a level of discourse structure that would predict the information structure of sentences within texts.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For the moment we will assume that there are only two possible labels: each y, is in { \u20141, +1}.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Also, expanding on the techniques for the automatic generation of extraction patterns (Riloff 96; Sudo 03) using our method, the extraction patterns which have the same meaning can be automatically linked, enabling us to produce the final table fully automatically.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Can we do . QmS: Yes, wonderful.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The problem is a binary classification problem.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"On the one hand, the type-level error rate is not calibrated for the number of n-grams in the sample.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As a partial solution, for pairs of hanzi that co-occur sufficiently often in our namelists, we use the estimated bigram cost, rather than the independence-based cost.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Among the IS-units, the referring expressions are marked as such and will in the second phase receive a label for cognitive status (active, accessible- text, accessible-situation, inferrable, inactive).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"However, we note that the final conditional estimates p(s|t) from a given phrase table maximize the likelihood of joint empirical phrase pair counts over a word-aligned corpus.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Since the transducers are built from human-readable descriptions using a lexical toolkit (Sproat 1995), the system is easily maintained and extended.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Our assumption is that caseframes that co-occur in resolutions often have a 2 This normalization is performed syntactically without semantics, so the agent and patient roles are not guaranteed to hold, but they usually do in practice.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We also mark all tags that dominate a word with the feminine ending :: taa mar buuTa (markFeminine).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"(2006) developed a technique for splitting and chunking long sentences.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The traveling salesman problem is an optimization problem which is defined as follows: given are a set of cities S = s1; ; sn and for each pair of cities si; sj the cost dij > 0 for traveling from city si to city sj . We are looking for the shortest tour visiting all cities exactly once while starting and ending in city s1.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The ATB annotation distinguishes between verbal and nominal readings of maSdar process nominals.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Sheffield Guthrie and Hepple (2010) explore several randomized compression techniques, but did not release code.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"In this case we are interested in finding' the maximum probability parse, ri, and Mi is the set of relevant (binary) parsing decisions made by parser i. ri is a parse selected from among the outputs of the individual parsers.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Also there are cases where one of the two NEs belong to a phrase outside of the relation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This is an issue that we have not addressed at the current stage of our research.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The number of NE instance pairs used in their experiment is less than half of our method.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"To check for this, we do pairwise bootstrap resampling: Again, we repeatedly sample sets of sentences, this time from both systems, and compare their BLEU scores on these sets.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"As was explained in the results section, \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d are not desirable keywords in the CC-domain.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In the numerator, however, the counts of ni1s are quite irregular, in\u00c2\u00ad cluding several zeros (e.g., RAT, none of whose members were seen).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Overview of the method 2.2 Step by Step Algorithm.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"from the subset of the United Informatics corpus not used in the training of the models.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Roughly speaking, the new algorithm presented in this paper performs a similar search, but instead minimizes a bound on the number of (unlabeled) examples on which two classifiers disagree.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"47 78.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"If two phrases can be used to express the same relationship within an information extraction application (\u00e2\u0080\u009cscenario\u00e2\u0080\u009d), these two phrases are paraphrases.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We start with noun features since written Arabic contains a very high proportion of NPs.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The resulting algorithm has a complexity of O(n!).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Phrase tables were extracted from the IN and OUT training corpora (not the dev as was used for instance weighting models), and phrase pairs in the intersection of the IN and OUT phrase tables were used as positive examples, with two alternate definitions of negative examples: The classifier trained using the 2nd definition had higher accuracy on a development set.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We also report word type level accuracy, the fraction of word types assigned their majority tag (where the mapping between model state and tag is determined by greedy one-to-one mapping discussed above).5 For each language, we aggregate results in the following way: First, for each hyperparameter setting, evaluate three variants: The first model (1TW) only 4 Typically, the performance stabilizes after only 10 itera-.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Search t0 CPU time #search mWER Method [sec] error [%] QmS 0.0 0.07 108 42:6 1.0 0.13 85 37:8 2.5 0.35 44 36:6 5.0 1.92 4 34:6 10.0 10.6 0 34:5 IbmS 0.0 0.14 108 43:4 1.0 0.3 84 39:5 2.5 0.8 45 39:1 5.0 4.99 7 38:3 10.0 28.52 0 38:2 Table 6 shows example translations obtained by the three different approaches.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Any opinions, findings, conclusions, or recommendations expressed in this paper are those of the authors, and do not necessarily reflect the views of the funding organizations.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The data is sorted based on the frequency of the context (\u00e2\u0080\u009ca unit of\u00e2\u0080\u009d appeared 314 times in the corpus) and the NE pair instances appearing with that context are shown with their frequency (e.g. \u00e2\u0080\u009cNBC\u00e2\u0080\u009d and \u00e2\u0080\u009cGeneral Electric Co.\u00e2\u0080\u009d appeared 10 times with the context \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Person\u00e2\u0080\u009d domain, 618 keywords are found, and in the \u00e2\u0080\u009cCountry \u00e2\u0080\u0093 Country\u00e2\u0080\u009d domain, 303 keywords are found.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This is a rather important source of errors in name identifi\u00c2\u00ad cation, and it is not really possible to objectively evaluate a name recognition system without considering the main lexicon with which it is used.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Pairwise comparison is done using the sign test.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If is a number string (such as one, two, etc), then the feature NumberString is set to 1.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"we perform five runs with different random initialization of sampling state.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"This is not an oversight.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"As already pointed out in Section 2.4, current theories diverge not only on the number and definition of relations but also on apects of structure, i.e., whether a tree is sufficient as a representational device or general graphs are required (and if so, whether any restrictions can be placed on these graph\u00e2\u0080\u0099s structures \u00e2\u0080\u0094 cf.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"However, our full model takes advantage of word features not present in Grac\u00c2\u00b8a et al.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As can be seen, GR and this \"pared-down\" statistical method perform quite similarly, though the statistical method is still slightly better.16 AG clearly performs much less like humans than these methods, whereas the full statistical algorithm, including morphological derivatives and names, performs most closely to humans among the automatic methods.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"0 57.3 51.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Local features are features that are based on neighboring tokens, as well as the token itself.","label":1},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"With the additional assumption that (s, t) can be restricted to the support of co(s, t), this is equivalent to a \u201cflat\u201d alternative to (6) in which each non-zero co(s, t) is set to one.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Finally, we add \u00e2\u0080\u009cDT\u00e2\u0080\u009d to the tags for definite nouns and adjectives (Kulick et al., 2006).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The computation of Pfr1(c)1Mi M k (C)) has been sketched before in Equations 1 through 4.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"See Section 5.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Gather phrases using keywords Now, the keyword with the top TF\/ITF score is selected for each phrase.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Consequently, all three parsers prefer the nominal reading.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Hence we decided to restrict ourselves to only information from the same document.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Future work should also extend the approach to build a complete named entity extractor - a method that pulls proper names from text and then classifies them.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"GL is then used to parse the string tn1 ... tnk_1, where tni is a terminal corresponding to the lattice span between node ni and ni+1.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Sometimes, however, these beliefs can be contradictory.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Se ma nti c (a) filters candidate if its semantic tags d o n \u00e2\u0080\u0099 t i n t e r s e c t w i t h t h o s e o f t h e a n a p h o r .","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For each pair of judges consider one judge as the standard,.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"In the rhetorical tree, nuclearity information is then used to extract a \u00e2\u0080\u009ckernel tree\u00e2\u0080\u009d that supposedly represents the key information from which the summary can be generated (which in turn may involve co-reference information, as we want to avoid dangling pronouns in a summary).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Second, the reduced number of hidden variables and parameters dramatically speeds up learning and inference.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We then use linguistic and annotation insights to develop a manually annotated grammar for Arabic (\u00c2\u00a74).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"na me =>1 ha nzi fa mi ly 1 ha nzi gi ve n 4.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Formalisms such as the restricted indexed grammars (Gazdar, 1985) and members of the hierarchy of grammatical systems given by Weir (1987) have independent paths, but more complex path sets.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Initially, we planned to compare the semantic classes of an anaphor and a candidate and infer that they might be coreferent if their semantic classes intersected.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The major problem for our seg\u00c2\u00ad menter, as for all segmenters, remains the problem of unknown words (see Fung and Wu [1994]).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"For instance, by altering the emission distribution parameters, Johnson (2007) encourages the model to put most of the probability mass on few tags.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Using cn to denote the number of n-grams, total memory consumption of TRIE, in bits, is plus quantization tables, if used.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Conditioned on T , features of word types W are drawn.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Restrictions We compare our new approach with the word reordering used in the IBM translation approach (Berger et al., 1996).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The first thing to note is that projectivizing helps in itself, even if no encoding is used, as seen from the fact that the projective baseline outperforms the non-projective training condition by more than half a percentage point on attachment score, although the gain is much smaller with respect to exact match.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"An example is in (i), where the system fails to group t;,f;?\"$?t!: lin2yang2gang3 as a name, because all three hanzi can in principle be separate words (t;,f; lin2 'wood';?\"$ yang2 'ocean'; ?t!; gang3 'harbor').","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We focus on phrases which connect two Named Entities (NEs), and proceed in two stages.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.6 Co-reference.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"But Arabic contains a variety of linguistic phenomena unseen in English.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The Prague Dependency Treebank (PDT) consists of more than 1M words of newspaper text, annotated on three levels, the morphological, analytical and tectogrammatical levels (Haji\u02c7c, 1998).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Borth 2MUC data can be obtained from the Linguistic Data Consortium: http:\/\/www.ldc.upenn.edu 3Training data for IdentiFinder is actually given in words (i.e., 650K & 790K words), rather than tokens Table 6: Comparison of results for MUC7 wick (1999) reported using dictionaries of person first names, corporate names and suffixes, colleges and universities, dates and times, state abbreviations, and world regions.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The accuracy is calculated as the ratio of the number of paraphrases to the total number of phrases in the set.","label":0},{"source_text":"There is no global pruning.","target_text":"f;g denotes the empty set, where no source sentence position is covered.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"( b ) s u p p o r t s c a n d i d a t e i f s e l e c t e d s e m a n t i c t a g s m a t c h t h o s e o f t h e a n a p h o r . Le xic al computes degree of lexical overlap b e t w e e n t h e c a n d i d a t e a n d t h e a n a p h o r . Re cen cy computes the relative distance between the c a n d i d a t e a n d t h e a n a p h o r . Sy nR ole computes relative frequency with which the c a n d i d a t e \u00e2\u0080\u0099 s s y n t a c t i c r o l e o c c u r s i n r e s o l u t i o n s . Figure 4: General Knowledge Sources The Lexical KS returns 1 if the candidate and anaphor are identical, 0.5 if their head nouns match, and 0 otherwise.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Then each arc of D maps either from an element of H to an element of p, or from E-i.e., the empty string-to an element of P. More specifically, each word is represented in the dictionary as a sequence of arcs, starting from the initial state of D and labeled with an element 5 of Hxp, which is terminated with a weighted arc labeled with an element of Ex P. The weight represents the estimated cost (negative log probability) of the word.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"(2010) and the posterior regular- ization HMM of Grac\u00c2\u00b8a et al.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"One striking example is the error reduction for Spanish, which reduces error by 36.5% and 24.7% for the best and median settings respectively.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"the number of permutations carried out for the word reordering is given.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Its only purpose is 3 This follows since each \u00ce\u00b8t has St \u00e2\u0088\u0092 1 parameters and.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Various segmentation approaches were then compared with human performance: 1.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Pr(eI 1) is the language model of the target language, whereas Pr(fJ 1 jeI1) is the transla tion model.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Although we have not done any experiments on other languages, this way of using global features from a whole document should be applicable to other languages.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Domain adaptation is a common concern when optimizing empirical NLP applications.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"It was then tested on section 22 of the Treebank in conjunction with the other parsers.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Evaluating parsing results in our joint framework, as argued by Tsarfaty (2006), is not trivial under the joint disambiguation task, as the hypothesized yield need not coincide with the correct one.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"4.4 Chinese Personal Names.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Table 3: Dev set frequencies for the two most significant discourse markers in Arabic are skewed toward analysis as a conjunction.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"These alignment models are similar to the concept of hidden Markov models (HMM) in speech recognition.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In total 13,976 phrases are assigned to sets of phrases, and the accuracy on our evaluation data ranges from 65 to 99%, depending on the domain and the size of the sets.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The focus of this work is on building POS taggers for foreign languages, assuming that we have an English POS tagger and some parallel text between the two languages.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, the company may refer to Company X in one paragraph and Company Y in another.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We develop our POS induction model based on the feature-based HMM of Berg-Kirkpatrick et al. (2010).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The question of what soft function to pick, and how to design' algorithms which optimize it, is an open question, but appears to be a promising way of looking at the problem.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Table 8a shows that the best model recovers SBAR at only 71.0% F1.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Table 5 provides insight into the behavior of different models in terms of the tagging lexicon they generate.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Collisions between two keys in the table can be identified at model building time.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"It is also worth pointing out a connection with Daum\u00b4e\u2019s (2007) work that splits each feature into domain-specific and general copies.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"x EG, has agreed to be bought by H x EG, now owned by H x H to acquire EG x H\u00e2\u0080\u0099s agreement to buy EG Three of those phrases are actually paraphrases, but sometime there could be some noise; such as the second phrase above.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Label propagation is used to propagate these tags inwards and results in tag distributions for the middle word of each Italian trigram.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The 13 positions of the source sentence are processed in the order shown.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As we shall argue, the semantic class affiliation of a hanzi constitutes useful information in predicting its properties.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Each set is assigned two values: belief and plausibility.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The number of ATB n-grams also falls below the WSJ sample size as the largest WSJ sample appeared in only 162 corpus positions.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"3.2 The DempsterShafer Decision Model.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Here, the term frequency (TF) is the frequency of a word in the bag and the inverse term frequency (ITF) is the inverse of the log of the frequency in the entire corpus.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"As we shall see, most of the linked sets are paraphrases.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Subsets C of increasing cardinality c are processed.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"As is standard, we use a fixed constant K for the number of tagging states.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"10) and trained both EM and L-BFGS for 1000 iterations.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In IE, creating the patterns which express the requested scenario, e.g. \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d or \u00e2\u0080\u009ccorporate merger and acquisition\u00e2\u0080\u009d is regarded as the hardest task.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The effect of a second reference resolution classifier is not entirely the same as that of global features.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The index in this array is the vocabulary identifier.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The features are used to represent each example for the learning algorithm.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Given counts cn1 where e.g. c1 is the vocabulary size, total memory consumption, in bits, is Our PROBING data structure places all n-grams of the same order into a single giant hash table.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":", for A. T.&T. nonalpha.. .","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We offer a state function s(wn1) = wn\ufffd where substring wn\ufffd is guaranteed to extend (to the right) in the same way that wn1 does for purposes of language modeling.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Source sentence words are aligned with hypothesized target sentence words, where the choice of a new source word, which has not been aligned with a target word yet, is restricted1.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If it starts with a lower case letter, and contains both upper and lower case letters, then (mixedCaps, zone) is set to 1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"As indicated in Figure 1(c), apart from this correct analysis, there is also the analysis taking B ri4 as a word (e.g., a common abbreviation for Japan), along with X:\u00c2\u00a5 wen2zhangl 'essay\/ and f!!.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We map the ATB morphological analyses to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags for all experiments.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The record for wn1 stores the offset at which its extensions begin.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We call this approach parser switching.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Until now, all evaluations of Arabic parsing\u00e2\u0080\u0094including the experiments in the previous section\u00e2\u0080\u0094have assumed gold segmentation.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The addition of vertical markovization enables non-pruned models to outperform all previously reported re12Cohen and Smith (2007) make use of a parameter (\u03b1) which is tuned separately for each of the tasks.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"With a good hash function, collisions of the full 64bit hash are exceedingly rare: one in 266 billion queries for our baseline model will falsely find a key not present.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"First, it directly encodes linguistic intuitions about POS tag assignments: the model structure reflects the one-tag-per-word property, and a type- level tag prior captures the skew on tag assignments (e.g., there are fewer unique determiners than unique nouns).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We also report word type level accuracy, the fraction of word types assigned their majority tag (where the mapping between model state and tag is determined by greedy one-to-one mapping discussed above).5 For each language, we aggregate results in the following way: First, for each hyperparameter setting, evaluate three variants: The first model (1TW) only 4 Typically, the performance stabilizes after only 10 itera-.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Sentence pairs are the natural instances for SMT, but sentences often contain a mix of domain-specific and general language.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"A Stochastic Finite-State Word-Segmentation Algorithm for Chinese","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Jiang and Zhai (2007) suggest the following derivation, making use of the true OUT distribution po(s, t): where each fi(s, t) is a feature intended to charac- !0\u02c6 = argmax pf(s, t) log p\u03b8(s|t) (8) terize the usefulness of (s, t), weighted by Ai. \u03b8 s,t pf(s, t)po(s, t) log p\u03b8(s|t) The mixing parameters and feature weights (col- != argmax po (s, t) lectively 0) are optimized simultaneously using dev- \u03b8 s,t pf(s, t)co(s, t) log p\u03b8(s|t), set maximum likelihood as before: !\ufffdargmax po (s, t) ! \u03b8 s,t \ufffd\u02c6 = argmax \u02dcp(s, t) log p(s|t; 0).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"18 77.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We experiment with four values for each hyperparameter resulting in 16 (\u00ce\u00b1, \u00ce\u00b2) combinations: \u00ce\u00b1 \u00ce\u00b2 0.001, 0.01, 0.1, 1.0 0.01, 0.1, 1.0, 10 Iterations In each run, we performed 30 iterations of Gibbs sampling for the type assignment variables W .4 We use the final sample for evaluation.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The first method uses a similar algorithm to that of (Yarowsky 95), with modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Time includes all queries but excludes random number generation and data structure population.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The test accuracy more or less asymptotes.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"If we remove this sample from the evaluation, then the ATB type-level error rises to only 37.4% while the n-gram error rate increases to 6.24%.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"There, a lattice is used to represent the possible sentences resulting from an interpretation of an acoustic model.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In practice, we can therefore expect a trade-off such that increasing the amount of information encoded in arc labels will cause an increase in the accuracy of the inverse transformation but a decrease in the accuracy with which the parser can construct the labeled representations.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"So, it is too costly to make IE technology \u00e2\u0080\u009copen- domain\u00e2\u0080\u009d or \u00e2\u0080\u009con-demand\u00e2\u0080\u009d like IR or QA.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Each production can push or pop symbols on the stack as can be seen in the following productions that generate tree of the form shown in Figure 4b.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"As such, global information from the whole context of a document is important to more accurately recognize named entities.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Other packages walk their respective data structures once to find wnf and again to find {b(wn\u22121 i )}f\u22121 i=1if necessary.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In Section 6 we dis\u00c2\u00ad cuss other issues relating to how higher-order language models could be incorporated into the model.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The perplexity for the trigram language model used is 26:5.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For all lists except locations, the lists are processed into a list of tokens (unigrams).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Similar advances have been made in machine translation (Frederking and Nirenburg, 1994), speech recognition (Fiscus, 1997) and named entity recognition (Borthwick et al., 1998).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"He used an additional maximum entropy classifier that tries to correct mistakes by using reference resolution.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For the translation model Pr(fJ 1 jeI 1), we go on the assumption that each source word is aligned to exactly one target word.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Following the convention presented in earlier sections, we assume that each example is an instance pair of the from (xi ,i, x2,) where xj,, E 2x3 , j E 2}.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Models that employ this strategy are denoted hsp.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"With respect to exact match, the improvement is even more noticeable, which shows quite clearly that even if non-projective dependencies are rare on the token level, they are nevertheless important for getting the global syntactic structure correct.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Considerable amount of work has been done in recent years on the named entity recognition task, partly due to the Message Understanding Conferences (MUC).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The hyperparameters \u00ce\u00b1 and \u00ce\u00b2 represent the concentration parameters of the token- and type-level components of the model respectively.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"A number of PCC commentaries will be read by professional news speakers and prosodic features be annotated, so that the various annotation layers can be set into correspondence with intonation patterns.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Here we propose a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Parameter Component As in the standard Bayesian HMM (Goldwater and Griffiths, 2007), all distributions are independently drawn from symmetric Dirichlet distributions: 2 Note that t and w denote tag and word sequences respectively, rather than individual tokens or tags.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"If two phrases can be used to express the same relationship within an information extraction application (\u00e2\u0080\u009cscenario\u00e2\u0080\u009d), these two phrases are paraphrases.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Better grammars are shown here to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"This approach is similar to BABAR in that they both acquire knowledge from earlier resolutions.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Tokens were converted to vocabulary identifiers in advance and state was carried from each query to the next.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The difference in performance between pronouns and definite noun phrases surprised us.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Sometimes, however, these beliefs can be contradictory.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"One may argue with these efforts on normalization, and ultimately their value should be assessed by assessing their impact on inter-annotator agreement.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"These sequence models-based approaches commonly treat token-level tag assignment as the primary latent variable.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The system of Berg-Kirkpatrick et al.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For the moment we will assume that there are only two possible labels: each y, is in { \u20141, +1}.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The developer explained that the loading process requires extra memory that it then frees. eBased on the ratio to SRI\u2019s speed reported in Guthrie and Hepple (2010) under different conditions.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"BerkeleyLM revision 152 (Pauls and Klein, 2011) implements tries based on hash tables and sorted arrays in Java with lossy quantization.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The web-based Annis imports data in a variety of XML formats and tagsets and displays it in a tier-orientedway (optionally, trees can be drawn more ele gantly in a separate window).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"For example, we can easily imagine that the number of paraphrases for \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d is enormous and it is not possible to create a comprehensive inventory by hand.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The input to AdaBoost is a set of training examples ((xi , yi), , (x\u201e.\u201e yrn)).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The 1st block contains the simple baselines from section 2.1.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Such a classification can be seen as a not-always-correct summary of global features.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Later, BerkeleyLM (Pauls and Klein, 2011) described ideas similar to ours.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"to explore how well we can induce POS tags using only the one-tag-per-word constraint.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"As was explained in the results section, \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d are not desirable keywords in the CC-domain.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"By contrast, when we turn to a comparison of the three encoding schemes it is hard to find any significant differences, and the overall impression is that it makes little or no difference which encoding scheme is used, as long as there is some indication of which words are assigned their linear head instead of their syntactic head by the projective parser.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"set was based on an earlier version of the Chang et a!.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J) and the lexicon probability p(fbi jei) are obtained by relative frequency estimates from the Viterbi alignment path after the final training iteration.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"and \u00e2\u0080\u009cH\u00e2\u0080\u009d represents \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"German city names are replaced by category markers.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For locations, tokens are matched against unigrams, and sequences of two consecutive tokens are matched against bigrams.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We used the MUC4 terrorism corpus (MUC4 Proceedings, 1992) and news articles from the Reuter\u00e2\u0080\u0099s text collection8 that had a subject code corresponding to natural disasters.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"When using the segmentation pruning (using HSPELL) for unseen tokens, performance improves for all tasks as well.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Each connected path (l1 ... lk) E L corresponds to one morphological segmentation possibility of W. The Parser Given a sequence of input tokens W = w1 ... wn and a morphological analyzer, we look for the most probable parse tree \u03c0 s.t.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"basically complete, yet some improvements and extensions are still under way.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Presence of the determiner J Al. 2.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Since our graph is built from a parallel corpus, we can use standard word alignment techniques to align the English sentences De 5Note that many combinations are impossible giving a PMI value of 0; e.g., when the trigram type and the feature instantiation don\u2019t have words in common. and their foreign language translations Df.6 Label propagation in the graph will provide coverage and high recall, and we therefore extract only intersected high-confidence (> 0.9) alignments De\ufffdf.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"47 78.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Consider the following sentences: (a) Jose Maria Martinez, Roberto Lisandy, and Dino Rossy, who were staying at a Tecun Uman hotel, were kidnapped by armed men who took them to an unknown place.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"It is difficult to motivate these days why one ministry should be exempt from cutbacks \u00e2\u0080\u0094 at the expense of the others.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The baseline simply retains the original labels for all arcs, regardless of whether they have been lifted or not, and the number of distinct labels is therefore simply the number n of distinct dependency types.2 In the first encoding scheme, called Head, we use a new label d\u2191h for each lifted arc, where d is the dependency relation between the syntactic head and the dependent in the non-projective representation, and h is the dependency relation that the syntactic head has to its own head in the underlying structure.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"hanzi in the various name positions, derived from a million names.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The dev corpus was taken from the NIST05 evaluation set, augmented with some randomly-selected material reserved from the training set.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"These estimates are in turn combined linearly with relative-frequency estimates from an in-domain phrase table.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Nonetheless there is no alternative to Reiche\u00e2\u0080\u0099s plan.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"36 79.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The less favored reading may be selected in certain contexts, however; in the case of , for example, the nominal reading jiang4 will be selected if there is morphological information, such as a following plural affix ir, menD that renders the nominal reading likely, as we shall see in Section 4.3.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Source sentence words are aligned with hypothesized target sentence words, where the choice of a new source word, which has not been aligned with a target word yet, is restricted1.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Although the segmentation requirements for Arabic are not as extreme as those for Chinese, Arabic is written with certain cliticized prepositions, pronouns, and connectives connected to adjacent words.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"This pumping lemma states that if there is tree, t = t2t3t4t5, generated by a TAG G, such that its height is more than a predetermined bound k, then all trees of the form ti it tstt ts for each i > 0 will also generated by G. Similarly, for tree sets with independent paths and more complex path sets, tree pumping lemmas can be given.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Note that we check for , the word preceding the consecutive sequence of initCaps tokens, since person prefixes like Mr., Dr., etc are not part of person names, whereas corporate suffixes like Corp., Inc., etc are part of corporate names.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We also add an annotation for one-level iDafa (oneLevelIdafa) constructs since they make up more than 75% of the iDafa NPs in the ATB (Gabbard and Kulick, 2008).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If is seen infrequently during training (less than a small count), then will not be selected as a feature and all features in this group are set to 0.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Te rro ris m Na tur al Dis ast ers mu rde r of < NP > kill ed <p atie nt > <a ge nt > da ma ged wa s inj ure d in < NP > <a ge nt > rep ort ed <a ge nt > add ed <a ge nt > occ urr ed cau se of < NP > <a ge nt > stat ed <a ge nt > add ed <a ge nt > wr eak ed <a ge nt > cro sse d per pet rat ed <p atie nt > con de mn ed <p atie nt > dri ver of < NP > <a ge nt > car ryi ng Figure 1: Caseframe Network Examples Figure 1 shows examples of caseframes that co-occur in resolutions, both in the terrorism and natural disaster domains.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"This result suggests the benefit of using the automatic discovery method.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"They cluster NE instance pairs based on the words in the contexts using a bag- of-words method.","label":1},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Because we are interested in applying our techniques to languages for which no labeled resources are available, we paid particular attention to minimize the number of free parameters and used the same hyperparameters for all language pairs.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"), and thosethat begin with a verb (\u00ef\u00bf\u00bd ub..i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We show that noun-noun vs. discourse-level coordination ambiguity in Arabic is a significant source of parsing errors (Table 8c).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In the second and third translation examples, the IbmS word reordering performs worse than the QmS word reordering, since it can not take properly into account the word reordering due to the German verbgroup.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As can be seen, GR and this \"pared-down\" statistical method perform quite similarly, though the statistical method is still slightly better.16 AG clearly performs much less like humans than these methods, whereas the full statistical algorithm, including morphological derivatives and names, performs most closely to humans among the automatic methods.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The constituent voting and na\u00efve Bayes techniques are equivalent because the parameters learned in the training set did not sufficiently discriminate between the three parsers.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Now, for this application one might be tempted to simply bypass the segmentation problem and pronounce the text character-by-character.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Blum and Mitchell 98) offer a promising formulation of redundancy, also prove some results about how the use of unlabeled examples can help classification, and suggest an objective function when training with unlabeled examples.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"During testing, it is possible that the classifier produces a sequence of inadmissible classes (e.g., person begin followed by location unique).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"13.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"(Webber et al., 2003)).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We call this technique constituent voting.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For example, ... fraud related to work on a federally funded sewage plant in Georgia In this case, Georgia is extracted: the NP containing it is a complement to the preposition in; the PP headed by in modifies the NP a federally funded sewage plant, whose head is the singular noun plant.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In the case of adverbial reduplication illustrated in (3b) an adjective of the form AB is reduplicated as AABB.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For the purposes of EM, the &quot;observed&quot; data is {(xi, Ya\u2022 \u2022 \u2022 (xrn, Yrn), xfil, and the hidden data is {ym+i y}.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We consider two variants of Berg-Kirkpatrick et al.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Another technique for parse hybridization is to use a na\u00efve Bayes classifier to determine which constituents to include in the parse.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We excluded these from the evaluation as they can be easily identified with a list of days\/months.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"We will briefly discuss this point in Section 3.1.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Storing state therefore becomes a time-space tradeoff; for example, we store state with partial hypotheses in Moses but not with each phrase.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"And then there are decisions that systems typically hard-wire, because the linguistic motivation for making them is not well understood yet.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Lexicon and OOV Handling Our data-driven morphological-analyzer proposes analyses for unknown tokens as described in Section 5.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"7 Unlike Dickinson (2005), we strip traces and only con-.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"KS Function Ge nde r filters candidate if gender doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For example, suppose one is building a ITS system for Mandarin Chinese.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We train linear mixture models for conditional phrase pair probabilities over IN and OUT so as to maximize the likelihood of an empirical joint phrase-pair distribution extracted from a development set.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"And then there are decisions that systems typically hard-wire, because the linguistic motivation for making them is not well understood yet.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"For the experiments, we use a simple preprocessing step.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"On several languages, we report performance exceeding that of state-of-the art systems.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This model is easily incorporated into the segmenter by building a WFST restrict\u00c2\u00ad ing the names to the four licit types, with costs on the arcs for any particular name summing to an estimate of the cost of that name.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We suggest that in unlexicalized PCFGs the syntactic context may be explicitly modeled in the derivation probabilities.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We also mark all nodes that dominate an SVO configuration (containsSVO).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For example, if is found in the list of person first names, the feature PersonFirstName is set to 1.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Following the convention presented in earlier sections, we assume that each example is an instance pair of the from (xi ,i, x2,) where xj,, E 2x3 , j E 2}.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"It is a sequence of proper nouns within an NP; its last word Cooper is the head of the NP; and the NP has an appositive modifier (a vice president at S.&P.) whose head is a singular noun (president).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For queries, we uniformly sampled 10 million hits and 10 million misses.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"There has also been some work on adapting the word alignment model prior to phrase extraction (Civera and Juan, 2007; Wu et al., 2005), and on dynamically choosing a dev set (Xu et al., 2007).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Additionally, for a given coverage set, at most 250 different hypotheses are kept during the search process, and the number of different words to be hypothesized by a source word is limited.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Instead, we focused on phrases and set the frequency threshold to 2, and so were able to utilize a lot of phrases while minimizing noise.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"More details on the parsing algorithm can be found in Nivre (2003).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We attempt to formalize this notion in terms of the tee pumping lemma which can be used to show that a tee set does not have dependent paths.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Ties are rare in Bayes switching because the models are fine-grained \u2014 many estimated probabilities are involved in each decision.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Apart from MERT difficulties, a conceptual problem with log-linear combination is that it multiplies feature probabilities, essentially forcing different features to agree on high-scoring candidates.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In a more recent study than Chang et al., Wang, Li, and Chang (1992) propose a surname-driven, non-stochastic, rule-based system for identifying personal names.17 Wang, Li, and Chang also compare their performance with Chang et al.'s system.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For example, if FCC and Federal Communications Commission are both found in a document, then Federal has A begin set to 1, Communications has A continue set to 1, Commission has A end set to 1, and FCC has A unique set to 1.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"These methods demonstrated the benefits of incorporating linguistic features using a log-linear parameterization, but requires elaborate machinery for training.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"computing the recall of the other's judgments relative to this standard.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We concentrate on those sets.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"An analysis of nouns that occur in both the singular and the plural in our database reveals that there is indeed a slight but significant positive correlation-R2 = 0.20, p < 0.005; see Figure 6.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The search starts in hypothesis (f;g; 0) and ends in the hypotheses (f1; ; Jg; j), with j 2 f1; ; Jg.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Again, the idea is that having a picture of syntax, co-reference, and sentence-internal information structure at one\u00e2\u0080\u0099s disposal should aid in finding models of discourse structure that are more explanatory and can be empirically supported.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"One obvious application is information extraction.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"can expect famous names like Zhou Enlai's to be in many dictionaries, but names such as :fi lf;f; shi2jil-lin2, the name of the second author of this paper, will not be found in any dictionary.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"of Articles No.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The content does not necessarily reflect the views of the U.S. Government, and no official endorsement should be inferred.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Before explaining our method in detail, we present a brief overview in this subsection.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"(7) \u03c6 s,t This is a somewhat less direct objective than used by Matsoukas et al, who make an iterative approximation to expected TER.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"gaolxing4 'happy'=> F.i'JF.i'JJI!JI!","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"72 78.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Quite often, though, these directives fulfill the goal of increasing annotator agreement without in fact settling the theoretical question; i.e., the directives are clear but not always very well motivated.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Second, the reduced number of hidden variables and parameters dramatically speeds up learning and inference.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"1 61.7 37.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Precision.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"A high-level relation is agent, which relates an animate nominal to a predicate.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Then, token- level HMM emission parameters are drawn conditioned on these assignments such that each word is only allowed probability mass on a single assigned tag.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Still, for both human and automatic rhetorical analysis, connectives are the most important source of surface information.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Fortunately, we were able to obtain a copy of the full set of sentences from Chang et al. on which Wang, Li, and Chang tested their system, along with the output of their system.18 In what follows we will discuss all cases from this set where our performance on names differs from that of Wang, Li, and Chang.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"D o m ai n Li n k ac cu ra cy W N c o v e r a g e C C 7 3 . 3 % 2 \/ 1 1 P C 8 8 . 9 % 2 \/ 8 Table 2.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Sentence pairs are the natural instances for SMT, but sentences often contain a mix of domain-specific and general language.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Corpus Step 1 NE pair instances Step 2 Step 1.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"While the proportion of sentences containing non-projective dependencies is often 15\u201325%, the total proportion of non-projective arcs is normally only 1\u20132%.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The choice of the particular newspaper was motivated by the fact that the language used in a regional daily is somewhat simpler than that of papers read nationwide.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"(2009).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"A cell in the bottom row of the parse chart is required for each potential whitespace boundary.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"A Brief Introduction to the Chinese Writing System Most readers will undoubtedly be at least somewhat familiar with the nature of the Chinese writing system, but there are enough common misunderstandings that it is as well to spend a few paragraphs on properties of the Chinese script that will be relevant to topics discussed in this paper.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Although matching is done at the sentence level, this information is subsequently discarded when all matches are pooled.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Our analysis identifies three key factors driving our performance gain: 1) selecting a model structure which directly encodes tag sparsity, 2) a type-level prior on tag assignments, and 3) a straightforward na\u00c2\u00a8\u00c4\u00b1veBayes approach to incorporate features.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"shows some keywords with their scores.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For each co-occurrence relation (noun\/caseframe for CFLex, and caseframe\/caseframe for CFNet), BABAR computes its log-likelihood value and looks it up in the \u00cf\u00872 table to obtain a confidence level.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The new algorithm, which we call CoBoost, uses labeled and unlabeled data and builds two classifiers in parallel.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"They are set to fixed constants.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"However, to classify a token , while Borthwick uses tokens from to (from two tokens before to two tokens after ), we used only the tokens , , and . Even with local features alone, MENERGI outperforms MENE (Borthwick, 1999).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Sometime, multiple words are needed, like \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, \u00e2\u0080\u009cprime minister\u00e2\u0080\u009d or \u00e2\u0080\u009cpay for\u00e2\u0080\u009d (\u00e2\u0080\u009cpay\u00e2\u0080\u009d and \u00e2\u0080\u009cpay for\u00e2\u0080\u009d are different senses in the CC-domain).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The data is sorted based on the frequency of the context (\u00e2\u0080\u009ca unit of\u00e2\u0080\u009d appeared 314 times in the corpus) and the NE pair instances appearing with that context are shown with their frequency (e.g. \u00e2\u0080\u009cNBC\u00e2\u0080\u009d and \u00e2\u0080\u009cGeneral Electric Co.\u00e2\u0080\u009d appeared 10 times with the context \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"This design does not guarantee \u00e2\u0080\u009cstructural zeros,\u00e2\u0080\u009d but biases towards sparsity.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"To quantize, we use the binning method (Federico and Bertoldi, 2006) that sorts values, divides into equally sized bins, and averages within each bin.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"If an unlabeled vertex does not have a path to any labeled vertex, this term ensures that the converged marginal for this vertex will be uniform over all tags, allowing the middle word of such an unlabeled vertex to take on any of the possible tags.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Cur In order to ground such approaches in linguistic observation and description, a multi-level anno 10 For an exposition of the idea as applied to the task of text planning, see (Chiarcos, Stede 2004).","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Formally, we define dependency graphs as follows: 3.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Analogous to the sharing of stacks in IC's, Lexical-Functional Grammar's (LFG's) use the unification of unbounded hierarchical structures.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Borth 2MUC data can be obtained from the Linguistic Data Consortium: http:\/\/www.ldc.upenn.edu 3Training data for IdentiFinder is actually given in words (i.e., 650K & 790K words), rather than tokens Table 6: Comparison of results for MUC7 wick (1999) reported using dictionaries of person first names, corporate names and suffixes, colleges and universities, dates and times, state abbreviations, and world regions.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We do not experiment with models larger than physical memory in this paper because TPT is unreleased, factors such as disk speed are hard to replicate, and in such situations we recommend switching to a more compact representation, such as RandLM.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"But gold segmentation is not available in application settings, so a segmenter and parser are arranged in a pipeline.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"This number must be less than or equal to n \u00f4\u0080\u0080\u0080 1.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The normalized judgement per sentence is the raw judgement plus (0 minus average raw judgement for this judge on this sentence).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Only tokens with initCaps not found in commonWords are tested against each list in Table 2.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"During training, we treat as observed the language word types W as well as the token-level corpus w. We utilize Gibbs sampling to approximate our collapsed model posterior: P (T ,t|W , w, \u00ce\u00b1, \u00ce\u00b2) \u00e2\u0088\u009d P (T , t, W , w|\u00ce\u00b1, \u00ce\u00b2) 0.7 0.6 0.5 0.4 0.3 English Danish Dutch Germany Portuguese Spanish Swedish = P (T , t, W , w, \u00cf\u0088, \u00ce\u00b8, \u00cf\u0086, w|\u00ce\u00b1, \u00ce\u00b2)d\u00cf\u0088d\u00ce\u00b8d\u00cf\u0086 Note that given tag assignments T , there is only one setting of token-level tags t which has mass in the above posterior.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"As discussed in more detail in \u00a73, we use two types of vertices in our graph: on the foreign language side vertices correspond to trigram types, while the vertices on the English side are individual word types.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Evaluation Metrics We report three metrics to evaluate tagging performance.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Further, we report current resident memory and peak virtual memory because these are the most applicable statistics provided by the kernel.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Figure 3 Figure 1.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"For each pair we also record the context, i.e. the phrase between the two NEs (Step1).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We focus here instead on adapting the two most important features: the language model (LM), which estimates the probability p(wIh) of a target word w following an ngram h; and the translation models (TM) p(slt) and p(t1s), which give the probability of source phrase s translating to target phrase t, and vice versa.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"This is true of the widely used link grammar parser for English (Sleator and Temperley, 1993), which uses a dependency grammar of sorts, the probabilistic dependency parser of Eisner (1996), and more recently proposed deterministic dependency parsers (Yamada and Matsumoto, 2003; Nivre et al., 2004).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The development set contained 44088 constituents in 2416 sentences and the test set contained 30691 constituents in 1699 sentences.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Analysis of the data revealed that the contextual role knowledge is especially helpful for resolving pronouns because, in general, they are semantically weaker than definite NPs.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In order to train this error-correction model, he divided his training corpus into 5 portions of 20% each.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"They contain about 200M words (25M, 110M, 40M and 19M words, respectively).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Figure 1 shows an example.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We use two common techniques, hash tables and sorted arrays, describing each before the model that uses the technique.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"This line of work has been motivated by empirical findings that the standard EM-learned unsupervised HMM does not exhibit sufficient word tag sparsity.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Kollege.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"When using the segmentation pruning (using HSPELL) for unseen tokens, performance improves for all tasks as well.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"I \u00e2\u0080\u00a2 JAPANS :rl4 .\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\)\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"o\u00c2\u00b7'\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\:J\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7 \u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7'\\; . '.:: ..........0 6.51 9.51 : jj \/ JAPANESE OCTOPUS 10\u00c2\u00b728i\u00c2\u00a3 :_nc HOW SAY f B :rl4 :il: :wen2 t '- \u00e2\u0080\u00a2 :zhang!","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This larger corpus was kindly provided to us by United Informatics Inc., R.O.C. a set of initial estimates of the word frequencies.9 In this re-estimation procedure only the entries in the base dictionary were used: in other words, derived words not in the base dictionary and personal and foreign names were not used.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The confidence level is then used as the belief value for the knowledge source.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"0 55.3 34.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For example, the phrase \u00e2\u0080\u009c's New York-based trust unit,\u00e2\u0080\u009d is not a paraphrase of the other phrases in the \u00e2\u0080\u009cunit\u00e2\u0080\u009d set.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"If 7' is an initial tree we do not include an address on the left-hand side.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"summit Sharm (a) Al-Sheikh summit Sharm (b) DTNNP Al-Sheikh in a corpus position without a bracketing label, then we also add \u00e2\u0088\u0097n, NIL) to M. We call the set of unique n-grams with multiple labels in M the variation nuclei of C. Bracketing variation can result from either annotation errors or linguistic ambiguity.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The key point is that the second constraint can be remarkably powerful in reducing the complexity of the learning problem.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"So the success of the algorithm may well be due to its success in maximizing the number of unlabeled examples on which the two decision lists agree.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The cost of storing these averages, in bits, is Because there are comparatively few unigrams, we elected to store them byte-aligned and unquantized, making every query faster.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"As is standard, we use a fixed constant K for the number of tagging states.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"In (Reitter, Stede 2003) we went a different way and suggested URML5, an XML format for underspecifying rhetorical structure: a number of relations can be assigned instead of a single one, competing analyses can be represented with shared forests.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Segmental morphology Hebrew consists of seven particles m(\u201cfrom\u201d) f(\u201cwhen\u201d\/\u201cwho\u201d\/\u201cthat\u201d) h(\u201cthe\u201d) w(\u201cand\u201d) k(\u201clike\u201d) l(\u201cto\u201d) and b(\u201cin\u201d). which may never appear in isolation and must always attach as prefixes to the following open-class category item we refer to as stem.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Note that while the standard HMM, has O(K n) emission parameters, our model has O(n) effective parameters.3 Token Component Once HMM parameters (\u00cf\u0086, \u00ce\u00b8) have been drawn, the HMM generates a token-level corpus w in the standard way: P (w, t|\u00cf\u0086, \u00ce\u00b8) = P (T , W , \u00ce\u00b8, \u00cf\u0088, \u00cf\u0086, t, w|\u00ce\u00b1, \u00ce\u00b2) = P (T , W , \u00cf\u0088|\u00ce\u00b2) [Lexicon] \u00ef\u00a3\u00ab n n \u00ef\u00a3\u00ad (w,t)\u00e2\u0088\u0088(w,t) j \u00ef\u00a3\u00b6 P (tj |\u00cf\u0086tj\u00e2\u0088\u00921 )P (wj |tj , \u00ce\u00b8tj )\u00ef\u00a3\u00b8 P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) [Parameter] P (w, t|\u00cf\u0086, \u00ce\u00b8) [Token] We refer to the components on the right hand side as the lexicon, parameter, and token component respectively.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"In Equations 1 through 3 we develop the model for constructing our parse using na\u00efve Bayes classification.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The frequency of the Company \u00e2\u0080\u0093 Company domain ranks 11th with 35,567 examples.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The record for wn1 stores the offset at which its extensions begin.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"All language model queries issued by machine translation decoders follow a left-to-right pattern, starting with either the begin of sentence token or null context for mid-sentence fragments.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The traveling salesman problem is an optimization problem which is defined as follows: given are a set of cities S = s1; ; sn and for each pair of cities si; sj the cost dij > 0 for traveling from city si to city sj . We are looking for the shortest tour visiting all cities exactly once while starting and ending in city s1.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"This is in contrast to dependency treebanks, e.g.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Here, the term frequency (TF) is the frequency of a word in the bag and the inverse term frequency (ITF) is the inverse of the log of the frequency in the entire corpus.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This might be because our features are more comprehensive than those used by Borthwick.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"How do additional ambiguities caused by devocalization affect statistical learning?","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Queries take the form p(wn|wn\u22121 1 ) where wn1 is an n-gram.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We construct a mapping from all the space-delimited tokens seen in the training sentences to their corresponding analyses.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This &quot;default&quot; feature type has 100% coverage (it is seen on every example) but a low, baseline precision.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Several extensions of AdaBoost for multiclass problems have been suggested (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Some approaches depend upon some form of con\u00c2\u00ad straint satisfaction based on syntactic or semantic features (e.g., Yeh and Lee [1991], which uses a unification-based approach).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This is a standard adaptation problem for SMT.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"gaolxing4 'happy' => F.i'JF.i'J Jl!","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Chinese word segmentation can be viewed as a stochastic transduction problem.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"For each experiment we gave an nonparametric and a parametric technique for combining parsers.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"It is worth noting that the middle words of the Italian trigrams are nouns too, which exhibits the fact that the similarity metric connects types having the same syntactic category.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The most frequently occurring last words of organization names in cslist are compiled into a list of corporate suffixes, Corporate-Suffix- List.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The last four columns in Table 3 show the distribution of nonprojective arcs with respect to the number of lifts required.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Once HMM parameters (\u00ce\u00b8, \u00cf\u0086) are drawn, a token-level tag and word sequence, (t, w), is generated in the standard HMM fashion: a tag sequence t is generated from \u00cf\u0086.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"State will ultimately be used as context in a subsequent query.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"For queries, we uniformly sampled 10 million hits and 10 million misses.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Affix Pron Base category N found N missed (recall) N correct (precision) t,-,7 The second issue is that rare family names can be responsible for overgeneration, especially if these names are otherwise common as single-hanzi words.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"BABAR uses two methods to identify anaphors that can be easily and reliably resolved with their antecedent: lexical seeding and syntactic seeding.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"All the NE pair instances which co-occur separated by at most 4 chunks are collected along with information about their NE types and the phrase between the NEs (the \u00e2\u0080\u0098context\u00e2\u0080\u0099).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Unigram lookup is dense so we use an array of probability and backoff values.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"I = 1X21 N and N is a &quot;medium&quot; sized number so that it is feasible to collect 0(N) unlabeled examples.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The development of the very first Hebrew Treebank (Sima\u2019an et al., 2001) called for the exploration of general statistical parsing methods, but the application was at first limited.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"2 62.2 39.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"But for eign learners are often surprised by the verbless predications that are frequently used in Arabic.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The simplest version of the maximum matching algorithm effectively deals with ambiguity by ignoring it, since the method is guaranteed to produce only one segmentation.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In this case, Maury Cooper is extracted.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For each caseframe, BABAR collects the semantic classes associated with the head nouns of NPs that were extracted by the caseframe.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"But Rehbein and van Genabith (2007) showed that Evalb should not be used as an indication of real difference\u00e2\u0080\u0094 or similarity\u00e2\u0080\u0094between treebanks.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For novel texts, no lexicon that consists simply of a list of word entries will ever be entirely satisfactory, since the list will inevitably omit many constructions that should be considered words.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For example, if Barry is not in commonWords and is found in the list of person first names, then the feature PersonFirstName will be set to 1.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Combining multiple highly-accurate independent parsers yields promising results.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"10.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"So, who won the competition?","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"21 84.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Lazy mapping reduces memory requirements by loading pages from disk only as necessary.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Space- or punctuation-delimited * 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"then define the best segmentation to be the cheapest or best path in Id(I) o D* (i.e., Id(I) composed with the transitive closure of 0).6 Consider the abstract example illustrated in Figure 2.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"phrase (markContainsVerb).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"An analysis of nouns that occur in both the singular and the plural in our database reveals that there is indeed a slight but significant positive correlation-R2 = 0.20, p < 0.005; see Figure 6.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Ex: Mr. Bush disclosed the policy by reading it...","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"To eliminate such sequences, we define a transition probability between word classes to be equal to 1 if the sequence is admissible, and 0 otherwise.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The hyperparameters \u00ce\u00b1 and \u00ce\u00b2 represent the concentration parameters of the token- and type-level components of the model respectively.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We refer to (T , W ) as the lexicon of a language and \u00cf\u0088 for the parameters for their generation; \u00cf\u0088 depends on a single hyperparameter \u00ce\u00b2.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Given counts cn1 where e.g. c1 is the vocabulary size, total memory consumption, in bits, is Our PROBING data structure places all n-grams of the same order into a single giant hash table.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"6 Joint Segmentation and Parsing.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The derivation trees of a MCTAG are similar to those of a TAG.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"It also incorporates the Good-Turing method (Baayen 1989; Church and Gale 1991) in estimating the likelihoods of previously unseen con\u00c2\u00ad structions, including morphological derivatives and personal names.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For example, in MUC6, there are four zones (TXT, HL, DATELINE, DD).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(Riloff and Jones 99) was brought to our attention as we were preparing the final version of this paper.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This means that the rules in our grammar are of two kinds: (a) syntactic rules relating nonterminals to a sequence of non-terminals and\/or PoS tags, and (b) lexical rules relating PoS tags to lattice arcs (lexemes).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Our experiments consistently demonstrate that this model architecture yields substantial performance gains over more complex tagging counterparts.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"more frequently than is done in English.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"However, the overall percentage of non-projective arcs is less than 2% in PDT and less than 1% in DDT.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"2.2 Syntactic structure.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Comparison with state-of-the-art taggers For comparison we consider two unsupervised tag- gers: the HMM with log-linear features of Berg- Kirkpatrick et al.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"However, as we have noted, nothing inherent in the approach precludes incorporating higher-order constraints, provided they can be effectively modeled within a finite-state framework.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Table 2 shows the features used in the current version of the parser.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Unigrams also have 64-bit overhead for vocabulary lookup.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"It did 402 queries\/ms using 1.80 GB. cMemory use increased during scoring due to batch processing (MIT) or caching (Rand).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This group contains a large number of features (one for each token string present in the training data).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The natural baseline approach is to concatenate data from IN and OUT.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This means that the PCC cannot grow particularly quickly.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"We use a solution to this problem similar to the one presented in (Och et al., 1999), where target words are joined during training.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"It reads a file with a list of German connectives, and when a text is opened for annotation, it highlights all the words that show up in this list; these will be all the potential connectives.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"(c) After they blindfolded the men...","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"IRSTLM (Federico et al., 2008) is an open-source toolkit for building and querying language models.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The ATB annotation guidelines specify that proper nouns should be specified with a flat NP (a).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Finally, we show that in application settings, the absence of gold segmentation lowers parsing performance by 2\u00e2\u0080\u00935% F1.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In this paper, we have presented a new, e\u00c3\u0086cient DP-based search procedure for statistical machine translation.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"t(i).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Restrictions: Quasi-monotone Search The above search space is still too large to allow the translation of a medium length input sentence.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The first four affixes are so-called resultative affixes: they denote some prop\u00c2\u00ad erty of the resultant state of a verb, as in E7 wang4bu4-liao3 (forget-not-attain) 'cannot forget.'","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"German English Training: Sentences 58 073 Words 519 523 549 921 Words* 418 979 453 632 Vocabulary Size 7939 4648 Singletons 3454 1699 Test-147: Sentences 147 Words 1 968 2 173 Perplexity { 26:5 Table 4: Multi-reference word error rate (mWER) and subjective sentence error rate (SSER) for three different search procedures.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Except our own and MENE + reference resolution, the results in Table 6 are all official MUC7 results.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The natural language processing community is in the strong position of having many available approaches to solving some of its most fundamental problems.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"2.1 Inverted Alignments.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Section 4.1 explained that state s is stored by applications with partial hypotheses to determine when they can be recombined.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We use the default inference parameters.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"When a token fmnh is to be interpreted as the lexeme sequence f\/REL mnh\/VB, the analysis introduces two distinct entities, the relativizer f (\u201cthat\u201d) and the verb mnh (\u201ccounted\u201d), and not as the complex entity \u201cthat counted\u201d.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Hyperparameter settings are sorted according to the median one-to-one metric over runs.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The second stage links sets which involve the same pairs of individual NEs.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The original OUT counts co(s, t) are weighted by a logistic function w\u03bb(s, t): To motivate weighting joint OUT counts as in (6), we begin with the \u201cideal\u201d objective for setting multinomial phrase probabilities 0 = {p(s|t), dst}, which is the likelihood with respect to the true IN distribution pi(s, t).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"But Rehbein and van Genabith (2007) showed that Evalb should not be used as an indication of real difference\u00e2\u0080\u0094 or similarity\u00e2\u0080\u0094between treebanks.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This has the effect of randomly permuting vocabulary identifiers, meeting the requirements of interpolation search when vocabulary identifiers are used as keys.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In order to facilitate this task, we extend the set of arc labels to encode information about lifting operations.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Similarly, for all the LCFRS's, discussed in Section 2, we can define the relationship between a structure and the sequence of substrings it spans, and the effect of the composition operations on sequences of substrings.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"yu2 'fish.'","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"All annotations are done with specific tools and in XML; each layer has its own DTD.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(a) ;IE shi4 'be' => ;IE;IE shi4bu2-shi4 (be-not-be) 'is it?'","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The computing time is given in terms of CPU time per sentence (on a 450MHz PentiumIIIPC).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In fact, during the first rounds many of the predictions of Th., g2 are zero.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The use of global features has improved the performance on MUC6 test data from 90.75% to 93.27% (27% reduction in errors), and the performance on MUC7 test data from 85.22% to 87.24% (14% reduction in errors).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"As one can see, only the trigrams [suo incarceramento ,], [suo iter ,] and [suo carattere ,] are connected to English words.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"In most cases, however, these expansions come with a steep increase in model complexity, with respect to training procedure and inference time.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We compare the manually annotated grammar, which we incorporate into the Stanford parser, to both the Berkeley (Petrov et al., 2006) and Bikel (Bikel, 2004) parsers.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The search starts in hypothesis (f;g; 0) and ends in the hypotheses (f1; ; Jg; j), with j 2 f1; ; Jg.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"There is a fairly large body of work on SMT adaptation.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The set of n-grams appearing in a model is sparse, and we want to efficiently find their associated probabilities and backoff penalties.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Vocabulary lookup is a sorted array of 64-bit word hashes.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Test set OOV rate is computed using the following splits: ATB (Chiang et al., 2006); CTB6 (Huang and Harper, 2009); Negra (Dubey and Keller, 2003); English, sections 221 (train) and section 23 (test).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Then, it can be verified that We can now derive the CoBoost algorithm as a means of minimizing Zco.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kinds of cue.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Such analyzers propose multiple segmentation possibilities and their corresponding analyses for a token in isolation but have no means to determine the most likely ones.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Two issues distinguish the various proposals.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The phrases have to be the expressions of length less than 5 chunks, appear between two NEs.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"(2009).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Our assumption is that caseframes that co-occur in resolutions often have a 2 This normalization is performed syntactically without semantics, so the agent and patient roles are not guaranteed to hold, but they usually do in practice.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The focus of this work is on building POS taggers for foreign languages, assuming that we have an English POS tagger and some parallel text between the two languages.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"For t = 1, T and for j = 1, 2: where 4 = exp(-jg'(xj,i)). practice, this greedy approach almost always results in an overall decrease in the value of Zco.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"BerkeleyLM revision 152 (Pauls and Klein, 2011) implements tries based on hash tables and sorted arrays in Java with lossy quantization.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Hence, the terminal symbols appearing in the structures that are composed are not lost (though a constant number of new symbols may be introduced).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(Riloff and Shepherd 97) describe a bootstrapping approach for acquiring nouns in particular categories (such as &quot;vehicle&quot; or &quot;weapon&quot; categories).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"ogy (Koskenniemi 1983; Antworth 1990; Tzoukermann and Liberman 1990; Karttunen, Kaplan, and Zaenen 1992; Sproat 1992); we represent the fact that ir, attaches to nouns by allowing t:-transitions from the final states of all noun entries, to the initial state of the sub-WFST representing f,.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"As with lexical expections, the semantic classes of co-referring expressions are 4 They may not be perfectly substitutable, for example one NP may be more specific (e.g., \u00e2\u0080\u009che\u00e2\u0080\u009d vs. \u00e2\u0080\u009cJohn F. Kennedy\u00e2\u0080\u009d).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"If these sets do not overlap, then the words cannot be coreferent.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"D o m ai n # of ph ras es t o t a l p h r a s e s ac cu ra cy C C 7 o r m o r e 1 0 5 8 7 . 6 % 6 o r l e s s 1 0 6 6 7 . 0 % P C 7 o r m o r e 3 5 9 9 9 . 2 % 6 o r l e s s 2 5 5 6 5 . 1 % Table 1.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Compared to decoding, this task is cache-unfriendly in that repeated queries happen only as they naturally occur in text.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"0 57.3 51.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"As in boosting, the algorithm works in rounds.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"To optimize this function, we used L-BFGS, a quasi-Newton method (Liu and Nocedal, 1989).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"A compatible view is presented by Charniak et al. (1996) who consider the kind of probabilities a generative parser should get from a PoS tagger, and concludes that these should be P(w|t) \u201cand nothing fancier\u201d.3 In our setting, therefore, the Lattice is not used to induce a probability distribution on a linear context, but rather, it is used as a common-denominator of state-indexation of all segmentations possibilities of a surface form.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"An interesting observation is that while vertical markovization benefits all our models, its effect is less evident in Cohen and Smith.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The simplest version of the maximum matching algorithm effectively deals with ambiguity by ignoring it, since the method is guaranteed to produce only one segmentation.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We include a list of per-category results for selected phrasal labels, POS tags, and dependencies in Table 8.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"88,962 (spelling,context) pairs were extracted as training data.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"6.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"This step ensures that the most frequent terms for each domain are labeled (in case some of them are not in WordNet) and labeled with the sense most appropriate for the domain.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"of Tokens MENERGI 318 160,000 200 180,000 IdentiFinder \u00e2\u0080\u0093 650,000 \u00e2\u0080\u0093 790,000 MENE \u00e2\u0080\u0093 \u00e2\u0080\u0093 350 321,000 Table 4: Training Data MUC7 test accuracy.2 For MUC6, the reduction in error due to global features is 27%, and for MUC7,14%.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In considering the relationship between formalisms, we show that it is useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees. find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars On the basis of this observation, we describe a class of formalisms which we call Linear Context- Free Rewriting Systems, and show they are recognizable in polynomial time and generate only semilinear languages.","label":1},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"MENE without Proteus, however, did not do very well and only achieved an F measure of 84.22% (Borthwick, 1999).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Following the convention presented in earlier sections, we assume that each example is an instance pair of the from (xi ,i, x2,) where xj,, E 2x3 , j E 2}.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Figure 2: Order in which source positions are visited for the example given in Fig.1.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Section 5 covers relevant previous work on SMT adaptation, and section 6 concludes.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Such resources exist for Hebrew (Itai et al., 2006), but unfortunately use a tagging scheme which is incompatible with the one of the Hebrew Treebank.s For this reason, we use a data-driven morphological analyzer derived from the training data similar to (Cohen and Smith, 2007).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The operations must be linear and nonerasing, i.e., they can not duplicate or erase structure from their arguments.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"6. 3) all G o l d P O S 7 0 0.7 91 0.825 358 0.7 73 0.818 358 0.8 02 0.836 452 80.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"9 66.4 47.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"40 75.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The next two rows are results of oracle experiments.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"5.2 Discussion.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"buy - acquire (5) buy - agree (2) buy - purchase (5) buy - acquisition (7) buy - pay (2)* buy - buyout (3) buy - bid (2) acquire - purchase (2) acquire - acquisition (2) acquire - pay (2)* purchase - acquisition (4) purchase - stake (2)* acquisition - stake (2)* unit - subsidiary (2) unit - parent (5) It is clear that these links form two clusters which are mostly correct.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"To facilitate comparison with previous work, we exhaustively evaluate this grammar and two other parsing models when gold segmentation is assumed (\u00c2\u00a75).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Our empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Different sentence structure and rich target language morphology are two reasons for this.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"AdaBoost finds a weighted combination of simple (weak) classifiers, where the weights are chosen to minimize a function that bounds the classification error on a set of training examples.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Thus in an English sentence such as I'm going to show up at the ACL one would reasonably conjecture that there are eight words separated by seven spaces.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"As mentioned above, it is not obvious how to apply Daum\u00b4e\u2019s approach to multinomials, which do not have a mechanism for combining split features.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"More formally, we start by representing the dictionary D as a Weighted Finite State Trans\u00c2\u00ad ducer (WFST) (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The Hebrew token \u2018bcl\u20191, for example, stands for the complete prepositional phrase 'We adopt here the transliteration of (Sima\u2019an et al., 2001).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Vocabulary lookup is a hash table mapping from word to vocabulary index.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(3)) to be defined over unlabeled as well as labeled instances.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Finally, the statistical method fails to correctly group hanzi in cases where the individual hanzi comprising the name are listed in the dictionary as being relatively high-frequency single-hanzi words.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Considerable amount of work has been done in recent years on the named entity recognition task, partly due to the Message Understanding Conferences (MUC).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"35 76.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Thus we opted not to take the step of creating more precise written annotation guidelines (as (Carlson, Marcu 2001) did for English), which would then allow for measuring inter-annotator agreement.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, some caveats are in order in comparing this method (or any method) with other approaches to seg\u00c2\u00ad mentation reported in the literature.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This &quot;default&quot; feature type has 100% coverage (it is seen on every example) but a low, baseline precision.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"A graph D = (W, A) is well-formed iff it is acyclic and connected.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The most accurate characterization of Chinese writing is that it is morphosyllabic (DeFrancis 1984): each hanzi represents one morpheme lexically and semantically, and one syllable phonologi\u00c2\u00ad cally.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"11 www.ling.unipotsdam.de\/sfb\/projekt a3.php 12 This step was carried out in the course of the diploma thesis work of David Reitter (2003), which de serves special mention here.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Because many foreign word types are not aligned to an English word (see Table 3), and we do not run label propagation on the foreign side, we expect the projected information to have less coverage.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Note that the backoff model assumes that there is a positive correlation between the frequency of a singular noun and its plural.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Altun et al. (2005) proposed a technique that uses graph based similarity between labeled and unlabeled parts of structured data in a discriminative framework for semi-supervised learning.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The commentaries in PCC are all of roughly the same length, ranging from 8 to 10 sentences.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"These two principles guide experimentation in this framework, and together with the evaluation measures help us decide which specific type of substructure to combine.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"If a candidate has a belief value \u00e2\u0089\u00a5 .50, then we select that candidate as the antecedent for the anaphor.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, some caveats are in order in comparing this method (or any method) with other approaches to seg\u00c2\u00ad mentation reported in the literature.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The problem with these styles of evaluation is that, as we shall demonstrate, even human judges do not agree perfectly on how to segment a given text.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Linear weights are difficult to incorporate into the standard MERT procedure because they are \u201chidden\u201d within a top-level probability that represents the linear combination.1 Following previous work (Foster and Kuhn, 2007), we circumvent this problem by choosing weights to optimize corpus loglikelihood, which is roughly speaking the training criterion used by the LM and TM themselves.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"It is closer to the smaller value of precision and recall when there is a large skew in their values.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Finally, we make some improvements to baseline approaches.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Here we push the single-framework conjecture across the board and present a single model that performs morphological segmentation and syntactic disambiguation in a fully generative framework.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"To control for the effect of the HSPELL-based pruning, we also experimented with a morphological analyzer that does not perform this pruning.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The subtree under,; is excised from 7, the tree 7' is inserted in its place and the excised subtree is inserted below the foot of y'.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In the cotraining case, (Blum and Mitchell 98) argue that the task should be to induce functions Ii and f2 such that So Ii and 12 must (1) correctly classify the labeled examples, and (2) must agree with each other on the unlabeled examples.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Like verbs, maSdar takes arguments and assigns case to its objects, whereas it also demonstrates nominal characteristics by, e.g., taking determiners and heading iDafa (Fassi Fehri, 1993).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Corpus Step 1 NE pair instances Step 2 Step 1.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The probability distribution that satisfies the above property is the one with the highest entropy.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Nu mb er filters candidate if number doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Sometime, multiple words are needed, like \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, \u00e2\u0080\u009cprime minister\u00e2\u0080\u009d or \u00e2\u0080\u009cpay for\u00e2\u0080\u009d (\u00e2\u0080\u009cpay\u00e2\u0080\u009d and \u00e2\u0080\u009cpay for\u00e2\u0080\u009d are different senses in the CC-domain).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We would like to thank Eugene Charniak, Michael Collins, and Adwait Ratnaparkhi for enabling all of this research by providing us with their parsers and helpful comments.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The performance of our system on those sentences ap\u00c2\u00ad peared rather better than theirs.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Further, Maamouri and Bies (2004) argued that the English guidelines generalize well to other languages.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This is not unreasonable given the application to phrase pairs from OUT, but it suggests that an interesting alternative might be to use a plain log-linear weighting function exp(Ei Aifi(s, t)), with outputs in [0, oo].","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"6 Joint Segmentation and Parsing.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In principle, it would be possible to encode the exact position of the syntactic head in the label of the arc from the linear head, but this would give a potentially infinite set of arc labels and would make the training of the parser very hard.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"And then there are decisions that systems typically hard-wire, because the linguistic motivation for making them is not well understood yet.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Here we use a slightly different notion of lift, applying to individual arcs and moving their head upwards one step at a time: Intuitively, lifting an arc makes the word wk dependent on the head wi of its original head wj (which is unique in a well-formed dependency graph), unless wj is a root in which case the operation is undefined (but then wj \u2014* wk is necessarily projective if the dependency graph is well-formed).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"An input ABCD can be represented as an FSA as shown in Figure 2(b).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"In line with perplexity results from Table 1, the PROBING model is the fastest followed by TRIE, and subsequently other packages.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"2.6 Co-reference.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"To approximate these baselines, we implemented a very simple sentence selection algorithm in which parallel sentence pairs from OUT are ranked by the perplexity of their target half according to the IN language model.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our work is motivated by the observation that contextual roles can be critically important in determining the referent of a noun phrase.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Formally, we define dependency graphs as follows: 3.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For novel texts, no lexicon that consists simply of a list of word entries will ever be entirely satisfactory, since the list will inevitably omit many constructions that should be considered words.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We are not claiming that this method is almighty.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Exposing this information to the decoder will lead to better hypothesis recombination.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In this way we restrict the parameterization of a Language Original case English Danish Dutch German Spanish Swedish Portuguese 94.6 96.3 96.6 95.5 95.4 93.3 95.6 Table 1: Upper bound on tagging accuracy assuming each word type is assigned to majority POS tag.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"However, to classify a token , while Borthwick uses tokens from to (from two tokens before to two tokens after ), we used only the tokens , , and . Even with local features alone, MENERGI outperforms MENE (Borthwick, 1999).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We report the F1 value of both measures.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"so that 'door' would be and in this case the hanzi 7C, does not represent a syllable.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The system of Berg-Kirkpatrick et al.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"IRSTLM and BerkeleyLM use this state function (and a limit of N \u22121 words), but it is more strict than necessary, so decoders using these packages will miss some recombination opportunities.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The table shows that the lexicon tag frequency predicated by our full model are the closest to the gold standard.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"(Kehler, 1997) also used a DempsterShafer model to merge evidence from different sources for template-level coreference.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Interestingly, Chang et al. report 80.67% recall and 91.87% precision on an 11,000 word corpus: seemingly, our system finds as many names as their system, but with four times as many false hits.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"All our results are obtained by using only the official training data provided by the MUC conferences.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"But the city name Sharm Al- Sheikh is also iDafa, hence the possibility for the incorrect annotation in (b).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The Berkeley parser gives state-of-the-art performance for all metrics.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The problem with these styles of evaluation is that, as we shall demonstrate, even human judges do not agree perfectly on how to segment a given text.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Here, annotation proceeds in two phases: first, the domains and the units of IS are marked as such.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Arguably this consists of about three phonological words.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"There is a (costless) transition between the NC node and f,.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The ATB segmentation scheme is one of many alternatives.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In the CC-domain, there are 32 sets of phrases which contain more than 2 phrases.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Ex: He was found in San Jose, where ...","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The record for wn1 stores the offset at which its extensions begin.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"36.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The following recursive equation is evaluated: Qe0 (e; S; C; j) = (2) = p(fj je) max \u00c3\u0086;e00 np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) max (S0;j0) (S0 ;Cnfjg;j0)!(S;C;j) j02Cnfjg Qe00 (e0; S0;C n fjg; j0)o: The search ends in the hypotheses (I; f1; ; Jg; j).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"These packages are further described in Section 3.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"4.4 Chinese Personal Names.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Memory mapping also allows the same model to be shared across processes on the same machine.","label":0},{"source_text":"Here we present two algorithms.","target_text":"This section describes AdaBoost, which is the basis for the CoBoost algorithm.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Restrictions We compare our new approach with the word reordering used in the IBM translation approach (Berger et al., 1996).","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"(Again, the goal of also in structural features.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The ability to redistribute belief values across sets rather than individual hypotheses is key.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"This result suggests the benefit of using the automatic discovery method.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"It is a sequence of proper nouns within an NP; its last word Cooper is the head of the NP; and the NP has an appositive modifier (a vice president at S.&P.) whose head is a singular noun (president).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"On MUC6 data, MENERGI also achieves performance comparable to IdentiFinder when trained on similar amount of training data.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"However, 1 http:\/\/maxent.sourceforge.net 3.2 Testing.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We further thank Dr. J.-S.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The fraction of buckets that are empty is m\u22121 m , so average lookup time is O( m 1) and, crucially, constant in the number of entries.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"computing the precision of the other's judgments relative to this standard.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We carried out translation experiments in two different settings.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"However, the characterization given in the main body of the text is correct sufficiently often to be useful.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The paper explains the design decisions taken in the annotations, and describes a number of applications using this corpus with its multi-layer annotation.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Applying the function PROJECTIVIZE to the graph in Figure 1 yields the graph in Figure 2, where the problematic arc pointing to Z has been lifted from the original head jedna to the ancestor je.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We use two different similarity functions to define the edge weights among the foreign vertices and between vertices from different languages.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We combined evidence from four contextual role knowledge sources with evidence from seven general knowledge sources using a DempsterShafer probabilistic model.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Similarly, for all the LCFRS's, discussed in Section 2, we can define the relationship between a structure and the sequence of substrings it spans, and the effect of the composition operations on sequences of substrings.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"This algorithm can be applied to statistical machine translation.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Using the terminology of Kahane et al. (1998), we say that jedna is the syntactic head of Z, while je is its linear head in the projectivized representation.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Our assumption is that caseframes that co-occur in resolutions often have a 2 This normalization is performed syntactically without semantics, so the agent and patient roles are not guaranteed to hold, but they usually do in practice.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Frequency is calculated by counting the number of distinct previous tokens that each token has (e.g., if Electric Corp. is seen 3 times, and Manufacturing Corp. is seen 5 times during training, and Corp. is not seen with any other preceding tokens, then the \u00e2\u0080\u009cfrequency\u00e2\u0080\u009d of Corp. is 2).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In a more recent study than Chang et al., Wang, Li, and Chang (1992) propose a surname-driven, non-stochastic, rule-based system for identifying personal names.17 Wang, Li, and Chang also compare their performance with Chang et al.'s system.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Each extraction pattern represents a linguistic expression and a syntactic position indicating where a role filler can be found.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"7 www.cis.upenn.edu\/\u00e2\u0088\u00bcpdtb\/ 8 www.eml-research.de\/english\/Research\/NLP\/ Downloads had to buy a new car.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"To explicitly handle the word reordering between words in source and target language, we use the concept of the so-called inverted alignments as given in (Ney et al., 2000).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The reason why we did not train with both MUC6 and MUC7 training data at the same time is because the task specifications for the two tasks are not identical.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Specifically, the lexicon is generated as: P (T , W |\u00cf\u0088) =P (T )P (W |T ) Word Type Features (FEATS): Past unsupervised POS work have derived benefits from features on word types, such as suffix and capitalization features (Hasan and Ng, 2009; Berg-Kirkpatrick et al.,2010).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Table 3: Training and test conditions for the Verbmobil task (*number of words without punctuation marks).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Acronyms (ACRO): Words made up of all capitalized letters in the text zone will be stored as acronyms (e.g., IBM).","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"In the small window on the left, search queries can be entered, here one for an NP that has been annotated on the co-reference layer as bridging.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We used section 23 as the development set for our combining techniques, and section 22 only for final testing.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Morphological disambiguators that consider a token in context (an utterance) and propose the most likely morphological analysis of an utterance (including segmentation) were presented by Bar-Haim et al. (2005), Adler and Elhadad (2006), Shacham and Wintner (2007), and achieved good results (the best segmentation result so far is around 98%).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"For example, we can easily imagine that the number of paraphrases for \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d is enormous and it is not possible to create a comprehensive inventory by hand.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(e.g., N.Y. would contribute this feature, IBM would not). nonalpha=x Appears if the spelling contains any characters other than upper or lower case letters.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Overall, language modeling significantly impacts decoder performance.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"On several languages, we report performance exceeding that of more complex state-of-the art systems.1","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"If it starts with a lower case letter, and contains both upper and lower case letters, then (mixedCaps, zone) is set to 1.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Rather we believe several methods have to be developed using different heuristics to discover wider variety of paraphrases.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"RandLM is the clear winner in RAM utilization, but is also slower and lower quality.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Many hanzi have more than one pronunciation, where the correct.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Our results suggest that current parsing models would benefit from better annotation consistency and enriched annotation in certain syntactic configurations.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Moreover, they are used as substantives much 2 Unlike machine translation, constituency parsing is not significantly affected by variable word order.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In addition, each feature function is a binary function.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This is a unique object for which we are able to define a proper probability model.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In that work, mutual information was used to decide whether to group adjacent hanzi into two-hanzi words.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Presence of the determiner J Al. 2.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"As a partial solution, for pairs of hanzi that co-occur sufficiently often in our namelists, we use the estimated bigram cost, rather than the independence-based cost.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Our TRIE implements the popular reverse trie, in which the last word of an n-gram is looked up first, as do SRILM, IRSTLM\u2019s inverted variant, and BerkeleyLM except for the scrolling variant.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For fair comparison, we have tabulated all results with the size of training data used (Table 5 and Table 6).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We first define &quot;pseudo-labels&quot;,-yt, as follows: = Yi t sign(g 0\\ 2\u2014 kx2,m < i < n Thus the first m labels are simply copied from the labeled examples, while the remaining (n \u2014 m) examples are taken as the current output of the second classifier.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"With some minor API changes, namely returning the length of the n-gram matched, it could also be faster\u2014though this would be at the expense of an optimization we explain in Section 4.1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"30 16.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Ablation Analysis We evaluate the impact of incorporating various linguistic features into our model in Table 3.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In fact, during the first rounds many of the predictions of Th., g2 are zero.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Can we do . QmS: Yes, wonderful.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Quite often, though, these directives fulfill the goal of increasing annotator agreement without in fact settling the theoretical question; i.e., the directives are clear but not always very well motivated.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Firstly, Hebrew unknown tokens are doubly unknown: each unknown token may correspond to several segmentation possibilities, and each segment in such sequences may be able to admit multiple PoS tags.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A summary of the corpus used in the experiments is given in Table 3.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"With regard to purely morphological phenomena, certain processes are not han\u00c2\u00ad dled elegantly within the current framework Any process involving reduplication, for instance, does not lend itself to modeling by finite-state techniques, since there is no way that finite-state networks can directly implement the copying operations required.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"(1), Pr(eI 1) is the language model, which is a trigram language model in this case.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank, leaving only sections 22 and 23 completely untouched during the development of any of the parsers.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"These two principles guide experimentation in this framework, and together with the evaluation measures help us decide which specific type of substructure to combine.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For the PROBING implementation, hash table sizes are in the millions, so the most relevant values are on the right size of the graph, where linear probing wins.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The probability distribution that satisfies the above property is the one with the highest entropy.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(We would like to note though that unlike previous boosting algorithms, the CoBoost algorithm presented here is not a boosting algorithm under Valiant's (Valiant 84) Probably Approximately Correct (PAC) model.)","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"IL+-1Proof: Assume a pair of crossing constituents appears in the output of the constituent voting technique using k parsers.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The following two error criteria are used in our experiments: mWER: multi-reference WER: We use the Levenshtein distance between the automatic translation and several reference translations as a measure of the translation errors.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Each learner is free to pick the labels for these instances.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"3.1 Corpora.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":", Sun day, then the feature DayOfTheWeek is set to 1.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"This feature has a linguistic justification.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Clearly, explicitly modeling such a powerful constraint on tagging assignment has a potential to significantly improve the accuracy of an unsupervised part-of-speech tagger learned without a tagging dictionary.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The edge weights between the foreign language trigrams are computed using a co-occurence based similarity function, designed to indicate how syntactically similar the middle words of the connected trigrams are (\u00a73.2).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The maximum precision row is the upper bound on accuracy if we could pick exactly the correct constituents from among the constituents suggested by the three parsers.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In this paper we present a stochastic finite-state model for segmenting Chinese text into words, both words found in a (static) lexicon as well as words derived via the above-mentioned productive processes.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"F-measure is the harmonic mean of precision and recall, 2PR\/(P + R).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Under certain conditions the constituent voting and na\u00efve Bayes constituent combination techniques are guaranteed to produce sets of constituents with no crossing brackets.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Instead of offsetting new topics with punctuation, writers of MSA in sert connectives such as \u00ef\u00bf\u00bd wa and \u00ef\u00bf\u00bd fa to link new elements to both preceding clauses and the text as a whole.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Our smoothing procedure takes into account all the aforementioned aspects and works as follows.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Time includes all queries but excludes random number generation and data structure population.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Their theoretical finding is simply stated: classification error rate decreases toward the noise rate exponentially in the number of independent, accurate classifiers.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This information is readily available in TRIE where adjacent records with equal pointers indicate no further extension of context is possible.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In this paper, we show how non-projective dependency parsing can be achieved by combining a datadriven projective parser with special graph transformation techniques.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The first concerns how to deal with ambiguities in segmentation.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"However, lexically similar NPs usually refer to the same entity in two cases: proper names and existential noun phrases.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In this work we extended the AdaBoost.MH (Schapire and Singer 98) algorithm to the cotraining case.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"This problem arises because our keywords consist of only one word.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"First of all, most previous articles report perfor\u00c2\u00ad mance in terms of a single percent-correct score, or else in terms of the paired measures of precision and recall.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"(2010) and the posterior regular- ization HMM of Grac\u00c2\u00b8a et al.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"More formally, we start by representing the dictionary D as a Weighted Finite State Trans\u00c2\u00ad ducer (WFST) (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In the denomi 11 We have two such lists, one containing about 17,000 full names, and another containing frequencies of.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The work reported here is closely related to [Ha- segawa et al. 04].","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Borth 2MUC data can be obtained from the Linguistic Data Consortium: http:\/\/www.ldc.upenn.edu 3Training data for IdentiFinder is actually given in words (i.e., 650K & 790K words), rather than tokens Table 6: Comparison of results for MUC7 wick (1999) reported using dictionaries of person first names, corporate names and suffixes, colleges and universities, dates and times, state abbreviations, and world regions.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"IRSTLM (Federico et al., 2008) is an open-source toolkit for building and querying language models.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(8) can now be rewritten5 as which is of the same form as the function Zt used in AdaBoost.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":1},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Sometimes extracted phrases by themselves are not meaningful to consider without context, but we set the following criteria.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In an experiment on automatic rhetorical parsing, the RST-annotations and PoS tags were used by (Reitter 2003) as a training corpus for statistical classification with Support Vector Machines.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Assume that the two classifiers are &quot;rote learners&quot;: that is, 1.1 and 12 are defined through look-up tables that list a label for each member of X1 or X2.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This is orthographically represented as 7C.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Cohen and Smith (2007) chose a metric like SParseval (Roark et al., 2006) that first aligns the trees and then penalizes segmentation errors with an edit-distance metric.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"To bridge this gap, we consider a practically motivated scenario, in which we want to leverage existing resources from a resource-rich language (like English) when building tools for resource-poor foreign languages.1 We assume that absolutely no labeled training data is available for the foreign language of interest, but that we have access to parallel data with a resource-rich language.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We further report SYNCS, the parsing metric of Cohen and Smith (2007), to facilitate the comparison.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"97 78.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"At this step, we will try to link those sets, and put them into a single cluster.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Therefore, performance is more closely tied to the underlying data structure than to the cache.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: He was found in San Jose, where ...","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"RandLM and SRILM also remove context that will not extend, but SRILM performs a second lookup in its trie whereas our approach has minimal additional cost.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"However, this result is consistent with the results of ex\u00c2\u00ad periments discussed in Wu and Fung (1994).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We often encounter sentences that are highly ambiguous in themselves, without some prior knowledge of the entities concerned.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"(If the TF\/IDF score of that word is below a threshold, the phrase is discarded.)","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"\u00e2\u0080\u009cAgree\u00e2\u0080\u009d is a subject control verb, which dominates another verb whose subject is the same as that of \u00e2\u0080\u009cagree\u00e2\u0080\u009d; the latter verb is generally the one of interest for extraction.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"3.2 Inter-annotator Agreement.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"As noted for the perplexity task, we do not expect cache to grow substantially with model size, so RandLM remains a low-memory option.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The index in this array is the vocabulary identifier.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Hash tables are a common sparse mapping technique used by SRILM\u2019s default and BerkeleyLM\u2019s hashed variant.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"It can be shown that this objective is convex in q.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"However, they list two sets, one consisting of 28 fragments and the other of 22 fragments, in which they had 0% recall and precision.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In contrast to these approaches, our method directly incorporates these constraints into the structure of the model.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"For example, we might have VP \u00e2\u0086\u0092 VB NP PP, where the NP is the subject.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If any of the tokens from to is in Person-Prefix- List, then another feature Person-Prefix is set to 1.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"By using the output of a hand-coded system such as Proteus, MENE can improve its performance, and can even outperform IdentiFinder (Borthwick, 1999).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"As we shall see, most of the linked sets are paraphrases.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This fact annoyed especially his dog...).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We report micro-averaged (whole corpus) and macro-averaged (per sentence) scores along add a constraint on the removal of punctuation, which has a single tag (PUNC) in the ATB.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"pre-processing.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The computing time is given in terms of CPU time per sentence (on a 450MHz PentiumIIIPC).","label":0},{"source_text":"There is no global pruning.","target_text":"What is important and is not expressed by the notation is the so-called coverage constraint: each source position j should be 'hit' exactly once by the path of the inverted alignment bI 1 = b1:::bi:::bI . Using the inverted alignments in the maximum approximation, we obtain as search criterion: max I (p(JjI) max eI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) max bI 1 I Yi=1 [p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])) = = max I (p(JjI) max eI 1;bI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])); where the two products over i have been merged into a single product over i. p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) is the trigram language model probability.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Cur In order to ground such approaches in linguistic observation and description, a multi-level anno 10 For an exposition of the idea as applied to the task of text planning, see (Chiarcos, Stede 2004).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"A cell in the bottom row of the parse chart is required for each potential whitespace boundary.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"our full model yields 39.3% average error reduction across languages when compared to the basic configuration (1TW).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"A total of 13,976 phrases were grouped.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The number of NE instance pairs used in their experiment is less than half of our method.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In Section 4, we present the performance measures used and give translation results on the Verbmobil task.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This information is readily available in TRIE where adjacent records with equal pointers indicate no further extension of context is possible.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, there is again local grammatical information that should favor the split in the case of (1a): both .ma3 'horse' and .ma3 lu4 are nouns, but only .ma3 is consistent with the classifier pil, the classifier for horses.21 By a similar argument, the preference for not splitting , lm could be strengthened in (lb) by the observation that the classifier 1'1* tiao2 is consistent with long or winding objects like , lm ma3lu4 'road' but not with,ma3 'horse.'","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We adopted this state-of-the-art model because it makes it easy to experiment with various ways of incorporating our novel constraint feature into the log-linear emission model.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We observe similar trends when using another measure \u00e2\u0080\u0093 type-level accuracy (defined as the fraction of words correctly assigned their majority tag), according to which La ng ua ge M etr ic B K 10 E M B K 10 L B F G S G 10 F EA T S B es t F EA T S M ed ia n E ng lis h 1 1 m 1 4 8 . 3 6 8 . 1 5 6 . 0 7 5 . 5 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 0 . 9 6 6 . 4 4 7 . 8 6 6 . 4 D an is h 1 1 m 1 4 2 . 3 6 6 . 7 4 2 . 6 5 8 . 0 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 2 . 1 6 1 . 2 4 3 . 2 6 0 . 7 D ut ch 1 1 m 1 5 3 . 7 6 7 . 0 5 5 . 1 6 4 . 7 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 6 . 4 6 9 . 0 5 1 . 5 6 7 . 3 Po rtu gu es e 1 1 m 1 5 0 . 8 7 5 . 3 4 3 . 2 7 4 . 8 44 .5 69 .2 6 4 . 1 7 4 . 5 5 6 . 5 7 0 . 1 S pa ni sh 1 1 m 1 \u00e2\u0080\u0093 \u00e2\u0080\u0093 4 0 . 6 7 3 . 2 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 8 . 3 6 8 . 9 5 0 . 0 5 7 . 2 Table 4: Comparison of our method (FEATS) to state-of-the-art methods.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The graph was constructed using 2 million trigrams; we chose these by truncating the parallel datasets up to the number of sentence pairs that contained 2 million trigrams.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"When a company buys another company, a paying event can occur, but these two phrases do not indicate the same event.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The size of TRIE is particularly sensitive to F1092 c11, so vocabulary filtering is quite effective at reducing model size.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Automatic Paraphrase Discovery based on Context and Keywords between NE Pairs","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"paper, and is missing 6 examples from the A set.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"IRSTLM 5.60.02 (Federico et al., 2008) is a sorted trie implementation designed for lower memory consumption.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The confidence intervals are computed by bootstrap resampling for BLEU, and by standard significance testing for the manual scores, as described earlier in the paper.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Supervised learning approaches have advanced the state-of-the-art on a variety of tasks in natural language processing, resulting in highly accurate systems.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Tsarfaty (2006) used a morphological analyzer (Segal, 2000), a PoS tagger (Bar-Haim et al., 2005), and a general purpose parser (Schmid, 2000) in an integrated framework in which morphological and syntactic components interact to share information, leading to improved performance on the joint task.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"For the experiments, we use a simple preprocessing step.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Finally, we make some improvements to baseline approaches.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In Table 5 we present results from small test cor\u00c2\u00ad pora for the productive affixes handled by the current version of the system; as with names, the segmentation of morphologically derived words is generally either right or wrong.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Two subjects are each given a calendar and they are asked to schedule a meeting.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(6), with W+ > W_.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Note that we check for , the word preceding the consecutive sequence of initCaps tokens, since person prefixes like Mr., Dr., etc are not part of person names, whereas corporate suffixes like Corp., Inc., etc are part of corporate names.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Word type N % Dic tion ary entr ies 2 , 5 4 3 9 7 . 4 7 Mor pho logi call y deri ved wor ds 3 0 . 1 1 Fore ign tran slite rati ons 9 0 . 3 4 Per son al na mes 5 4 2 . 0 7 cases.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We implement two data structures: PROBING, designed for speed, and TRIE, optimized for memory.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For derived words that occur in our corpus we can estimate these costs as we would the costs for an underived dictionary entry.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The features we used can be divided into 2 classes: local and global.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Find keywords for each NE pair When we look at the contexts for each domain, we noticed that there is one or a few important words which indicate the relation between the NEs (for example, the word \u00e2\u0080\u009cunit\u00e2\u0080\u009d for the phrase \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, if {N P1, N P2, N P3} are all coreferent, then each NP must be linked to one of the other two NPs.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Once the lexicon has been drawn, the model proceeds similarly to the standard token-level HMM: Emission parameters \u00ce\u00b8 are generated conditioned on tag assignments T . We also draw transition parameters \u00cf\u0086.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"2.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Clearly this poses a number of research challenges, though, such as the applicability of tag sets across different languages.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This makes memory usage comparable to our PROBING model.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Table 1: Syntactic Seeding Heuristics BABAR\u00e2\u0080\u0099s reliable case resolution heuristics produced a substantial set of anaphor\/antecedent resolutions that will be the training data used to learn contextual role knowledge.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The ATB annotation guidelines specify that proper nouns should be specified with a flat NP (a).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Also surprising is the low test set OOV rate given the possibility of morphological variation in Arabic.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Several methods exist to handle collisions; we use linear probing because it has less memory overhead when entries are small.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"This overview is illustrated in Figure 1.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Mandarin exhibits several such processes, including A-not-A question formation, il\u00c2\u00ad lustrated in (3a), and adverbial reduplication, illustrated in (3b): 3.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"5 Related Work.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Table 1 briefly describes the seven syntactic heuristics used by BABAR to resolve noun phrases.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Instead, we extend the variation n-gram method of Dickinson (2005) to compare annotation error rates in the WSJ and ATB.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We used it to score all phrase pairs in the OUT table, in order to provide a feature for the instance-weighting model.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"(2009).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"TAG's can be used to give the structural descriptions discussed by Gazdar (1985) for the unbounded nested dependencies in Norwedish, for cross serial dependencies in Dutch subordinate clauses, and for the nestings of paired English complementizers.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For instance, for TTS it is necessary to know that a particular sequence of hanzi is of a particular category because that knowl\u00c2\u00ad edge could affect the pronunciation; consider, for example the issues surrounding the pronunciation of ganl I qian2 discussed in Section 1.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"AdaBoost.MH maintains a distribution over instances and labels; in addition, each weak-hypothesis outputs a confidence vector with one confidence value for each possible label.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"keywords Step 3 Sets of phrases based on keywords Step 4 Links between sets of phrases All the contexts collected for a given domain are gathered in a bag and the TF\/ITF scores are calculated for all the words except stopwords in the bag.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The average agreement among the human judges is .76, and the average agreement between ST and the humans is .75, or about 99% of the interhuman agreement.15 One can better visualize the precision-recall similarity matrix by producing from that matrix a distance matrix, computing a classical metric multidimensional scaling (Torgerson 1958; Becker, Chambers, Wilks 1988) on that dis\u00c2\u00ad tance matrix, and plotting the first two most significant dimensions.","label":1},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The hypothesized parse is then the set of constituents that are likely (P > 0.5) to be in the parse according to this model.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"To initialize the graph we tag the English side of the parallel text using a supervised model.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The baseline simply retains the original labels for all arcs, regardless of whether they have been lifted or not, and the number of distinct labels is therefore simply the number n of distinct dependency types.2 In the first encoding scheme, called Head, we use a new label d\u2191h for each lifted arc, where d is the dependency relation between the syntactic head and the dependent in the non-projective representation, and h is the dependency relation that the syntactic head has to its own head in the underlying structure.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Table 4 Differences in performance between our system and Wang, Li, and Chang (1992).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For the moment we will assume that there are only two possible labels: each y, is in { \u20141, +1}.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"It is possible one could produce better models by introducing features describing constituents and their contexts because one parser could be much better than the majority of the others in particular situations.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"4 7 . 3 8 . 9 2 8 . 8 2 0 . 7 3 2 . 3 3 5 . 2 2 9 . 6 2 7 . 6 1 4 . 2 4 2 . 8 4 5 . 9 4 4 . 3 6 0 . 6 6 1 . 5 4 9 . 9 3 3 . 9 Table 6: Type-level Results: Each cell report the type- level accuracy computed against the most frequent tag of each word type.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"If one is interested in translation, one would probably want to consider show up as a single dictionary word since its semantic interpretation is not trivially derivable from the meanings of show and up.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The second row represents the performance of the median hyperparameter setting.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"There are still some open issues to be resolved with the format, but it represents a first step.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Figure 1 depicts the lattice for a 2-words sentence bclm hneim.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Benchmarks use the package\u2019s binary format; our code is also the fastest at building a binary file.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This group of features attempts to capture such information.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We train and test on the CoNLL-X training set.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"2 62.6 45.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Even with less training data, MENERGI outperforms Borthwick' s MENE + reference resolution (Borthwick, 1999).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In various dialects of Mandarin certain phonetic rules apply at the word.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thus, the language generated by a grammar of a LCFRS is semilinear.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"3.1 General Knowledge Sources.","label":0},{"source_text":"There is no global pruning.","target_text":"To formalize the approach, we introduce four verbgroup states S: Initial (I): A contiguous, initial block of source positions is covered.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The final score is obtained from: max e;e0 j2fJ\u00f4\u0080\u0080\u0080L;;Jg p($je; e0) Qe0 (e; I; f1; ; Jg; j); where p($je; e0) denotes the trigram language model, which predicts the sentence boundary $ at the end of the target sentence.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For example, from the sentence \u00e2\u0080\u009cMr.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We retain segmentation markers\u00e2\u0080\u0094which are consistent only in the vocalized section of the treebank\u00e2\u0080\u0094to differentiate between e.g. \u00ef\u00bf\u00bd \u00e2\u0080\u009cthey\u00e2\u0080\u009d and \u00ef\u00bf\u00bd + \u00e2\u0080\u009ctheir.\u00e2\u0080\u009d Because we use the vocalized section, we must remove null pronoun markers.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"A novel element of our model is the ability to capture type-level tag frequencies.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The dev and test sets were randomly chosen from the EMEA corpus.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Segments with the same surface form but different PoS tags are treated as different lexemes, and are represented as separate arcs (e.g. the two arcs labeled neim from node 6 to 7).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Here, $ is the sentence boundary symbol, which is thought to be at position 0 in the target sentence.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Our clue is the NE instance pairs.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"It then computes a normalized Levenshtein edit distance between the extracted chain and the reference.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"With a good hash function, collisions of the full 64bit hash are exceedingly rare: one in 266 billion queries for our baseline model will falsely find a key not present.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"There are clearly eight orthographic words in the example given, but if one were doing syntactic analysis one would probably want to consider I'm to consist of two syntactic words, namely I and am.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Affix Pron Base category N found N missed (recall) N correct (precision) t,-,7 The second issue is that rare family names can be responsible for overgeneration, especially if these names are otherwise common as single-hanzi words.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"computing the precision of the other's judgments relative to this standard.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Next, we describe four contextual role knowledge sources that are created from the training examples and the caseframes.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"A similar maximumlikelihood approach was used by Foster and Kuhn (2007), but for language models only.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Then, moving from connective to connective, ConAno sometimes offers suggestions for its scope (using heuristics like \u00e2\u0080\u0098for sub- junctor, mark all words up to the next comma as the first segment\u00e2\u0080\u0099), which the annotator can accept with a mouseclick or overwrite, marking instead the correct scope with the mouse.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"First, we aim to explicitly characterize examples from OUT as belonging to general language or not.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(We would like to note though that unlike previous boosting algorithms, the CoBoost algorithm presented here is not a boosting algorithm under Valiant's (Valiant 84) Probably Approximately Correct (PAC) model.)","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In this section, we describe how contextual role knowledge is represented and learned.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"segmentation (Table 2).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, we have reason to doubt Chang et al.'s performance claims.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Unsupervised Learning of Contextual Role Knowledge for Coreference Resolution","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"In particular, the decision to represent arguments in verb- initial clauses as VP internal makes VSO and VOS configurations difficult to distinguish.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"This family represents an attempt to generalize the properties shared by CFG's, HG's, TAG's, and MCTAG's.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"An input ABCD can be represented as an FSA as shown in Figure 2(b).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Table 6: Example Translations for the Verbmobil task.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"All experiments use ATB parts 1\u00e2\u0080\u00933 divided according to the canonical split suggested by Chiang et al.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Since foreign names can be of any length, and since their original pronunciation is effectively unlimited, the identi\u00c2\u00ad fication of such names is tricky.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The second model (+PRIOR) utilizes the independent prior over type-level tag assignments P (T |\u00cf\u0088).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Each trie node is individually allocated and full 64-bit pointers are used to find them, wasting memory.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Thus, provided at least this amount of IN data is available\u2014as it is in our setting\u2014adapting these weights is straightforward.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Since our graph is built from a parallel corpus, we can use standard word alignment techniques to align the English sentences De 5Note that many combinations are impossible giving a PMI value of 0; e.g., when the trigram type and the feature instantiation don\u2019t have words in common. and their foreign language translations Df.6 Label propagation in the graph will provide coverage and high recall, and we therefore extract only intersected high-confidence (> 0.9) alignments De\ufffdf.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Further, we report current resident memory and peak virtual memory because these are the most applicable statistics provided by the kernel.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The first is an evaluation of the system's ability to mimic humans at the task of segmenting text into word-sized units; the second evaluates the proper-name identification; the third measures the performance on morphological analysis.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"evaluated to account for the same fraction of the data.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Somewhat surprisingly, there do not appear to be large systematic differences between linear and MAP combinations.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The observed performance gains, coupled with the simplicity of model implementation, makes it a compelling alternative to existing more complex counterparts.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"A simple lexicalized PCFG with second order Markovization gives relatively poor performance: 75.95% F1 on the test set.8 But this figure is surprisingly competitive with a recent state-of-the-art baseline (Table 7).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"This locally normalized log-linear model can look at various aspects of the observation x, incorporating overlapping features of the observation.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"As an indication, in our core corpus, we found an average sentence length of 15.8 words and 1.8 verbs per sentence, whereas a randomly taken sample of ten commentaries from the national papers Su\u00c2\u00a8ddeutsche Zeitung and Frankfurter Allgemeine has 19.6 words and 2.1 verbs per sentence.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"On the other hand, if it is seen as McCann Pte.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This PP modifies another NP, whose head is a singular noun.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"For developers of Statistical Machine Translation (SMT) systems, an additional complication is the heterogeneous nature of SMT components (word-alignment model, language model, translation model, etc.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Virginia) and -sia are normally transliterated as fbSi!","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"( b ) s u p p o r t s c a n d i d a t e i f s e l e c t e d s e m a n t i c t a g s m a t c h t h o s e o f t h e a n a p h o r . Le xic al computes degree of lexical overlap b e t w e e n t h e c a n d i d a t e a n d t h e a n a p h o r . Re cen cy computes the relative distance between the c a n d i d a t e a n d t h e a n a p h o r . Sy nR ole computes relative frequency with which the c a n d i d a t e \u00e2\u0080\u0099 s s y n t a c t i c r o l e o c c u r s i n r e s o l u t i o n s . Figure 4: General Knowledge Sources The Lexical KS returns 1 if the candidate and anaphor are identical, 0.5 if their head nouns match, and 0 otherwise.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"A Single Generative Model for Joint Morphological Segmentation and Syntactic Parsing","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"When this metric is less than 0.5, we expect to incur more errors' than we will remove by adding those constituents to the parse.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"10.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We make the assumption that for each example, both xi,. and x2,2 alone are sufficient to determine the label yi.","label":0},{"source_text":"There is no global pruning.","target_text":"Fall.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Statistics are printed before Moses exits and after parts of the decoder have been destroyed.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Clearly, retaining the original frequencies is important for good performance, and globally smoothing the final weighted frequencies is crucial.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Our empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our assumption is that caseframes that co-occur in resolutions often have a 2 This normalization is performed syntactically without semantics, so the agent and patient roles are not guaranteed to hold, but they usually do in practice.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"So, it is too costly to make IE technology \u00e2\u0080\u009copen- domain\u00e2\u0080\u009d or \u00e2\u0080\u009con-demand\u00e2\u0080\u009d like IR or QA.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The zone to which a token belongs is used as a feature.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Various segmentation approaches were then compared with human performance: 1.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"These results are achieved by training on the official MUC6 and MUC7 training data, which is much less training data than is used by other machine learning systems that worked on the MUC6 or MUC7 named entity task (Bikel et al., 1997; Bikel et al., 1999; Borth- wick, 1999).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The advantage is that we can recombine search hypotheses by dynamic programming.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"From now on all lattice arcs are tagged segments and the assignment of probability P(p \u2014* (s, p)) to lattice arcs proceeds as usual.4 A rather pathological case is when our lexical heuristics prune away all segmentation possibilities and we remain with an empty lattice.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"At first glance, this seems only peripherally related to our work, since the specific\/general distinction is made for features rather than instances.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The model described here thus demonstrates great potential for use in widespread applications.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"However, those methods need initial seeds, so the relation between entities has to be known in advance.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Note that on some examples (around 2% of the test set) CoBoost abstained altogether; in these cases we labeled the test example with the baseline, organization, label.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"A Person-Prefix-List is compiled in an analogous way.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Each lattice arc corresponds to a segment and its corresponding PoS tag, and a path through the lattice corresponds to a specific morphological segmentation of the utterance.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Other packages walk their respective data structures once to find wnf and again to find {b(wn\u22121 i )}f\u22121 i=1if necessary.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"More details on the memory-based prediction can be found in Nivre et al. (2004) and Nivre and Scholz (2004).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"At this step, we will try to link those sets, and put them into a single cluster.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"4.1 Dictionary Representation.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Call the crossing constituents A and B.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Assigning rhetorical relations thus poses questions that can often be answered only subjectively.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Our second point of comparison is with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We create equivalence classes for verb, noun, and adjective POS categories.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We evaluate our model on seven languages exhibiting substantial syntactic variation.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Language models that contain wi must also contain prefixes wi for 1 G i G k. Therefore, when the model is queried for p(wnjwn\u22121 1 ) but the longest matching suffix is wnf , it may return state s(wn1) = wnf since no longer context will be found.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The dependency graph in Figure 1 satisfies all the defining conditions above, but it fails to satisfy the condition ofprojectivity (Kahane et al., 1998): The arc connecting the head jedna (one) to the dependent Z (out-of) spans the token je (is), which is not dominated by jedna.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Such ambiguities cause discrepancies between token boundaries (indexed as white spaces) and constituent boundaries (imposed by syntactic categories) with respect to a surface form.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"A spelling rule might be a simple look-up for the string (e.g., a rule that Honduras is a location) or a rule that looks at words within a string (e.g., a rule that any string containing Mr. is a person).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We call such a constituent an isolated constituent.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Model Overview The model starts by generating a tag assignment T for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"So far the graph has been completely unlabeled.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Secondly, for all our models we provide better fine- and coarse-grained POS-tagging accuracy, and all pruned models outperform the Oracle results reported by them.12 In terms of syntactic disambiguation, even the simplest grammar pruned with HSPELL outperforms their non-Oracle results.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The function A : F \u2014* C maps from the language specific fine-grained tagset F to the coarser universal tagset C and is described in detail in \u00a76.2: Note that when tx(y) = 1 the feature value is 0 and has no effect on the model, while its value is \u2212oc when tx(y) = 0 and constrains the HMM\u2019s state space.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The BerkeleyLM direct-mapped cache is in principle faster than caches implemented by RandLM and by IRSTLM, so we may write a C++ equivalent implementation as future work.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"During coreference resolution, BABAR checks (1) whether the anaphor is among the lexical expectations for the caseframe that extracts the candidate antecedent, and (2) whether the candidate is among the lexical expectations for the caseframe that extracts the anaphor.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This variant is tested in the experiments below.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Informally, two or more paths can be dependent on each other: for example, they could be required to be of equal length as in the trees in Figure 4. generates such a tree set.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Each word is terminated by an arc that represents the transduction between f and the part of speech of that word, weighted with an estimated cost for that word.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"(7) \u03c6 s,t This is a somewhat less direct objective than used by Matsoukas et al, who make an iterative approximation to expected TER.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We include a list of per-category results for selected phrasal labels, POS tags, and dependencies in Table 8.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Interpolation search is therefore a form of binary search with better estimates informed by the uniform key distribution.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This distributional sparsity of syntactic tags is not unique to English 1 The source code for the work presented in this paper is available at http:\/\/groups.csail.mit.edu\/rbg\/code\/typetagging\/.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Thus, rather than give a single evaluative score, we prefer to compare the performance of our method with the judgments of several human subjects.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"However there is no global pruning.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Process statistics are already collected by the kernel (and printing them has no meaningful impact on performance).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"In both cases the investigators were able to achieve significant improvements over the previous best tagging results.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"4.1 Corpora.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In contrast, even though context can be helpful for resolving definite NPs, context can be trumped by the semantics of the nouns themselves.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We insisted that each structure dominates a bounded number of (not necessarily adjacent) substrings.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Our test machine has two Intel Xeon E5410 processors totaling eight cores, 32 GB RAM, and four Seagate Barracuda disks in software RAID 0 running Linux 2.6.18.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Again, this deserves further investigation.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"This paper proposes a simple and effective tagging method that directly models tag sparsity and other distributional properties of valid POS tag assignments.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Also, in Information Extraction (IE), in which the system tries to extract elements of some events (e.g. date and company names of a corporate merger event), several event instances from different news articles have to be aligned even if these are expressed differently.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In practice, we can therefore expect a trade-off such that increasing the amount of information encoded in arc labels will cause an increase in the accuracy of the inverse transformation but a decrease in the accuracy with which the parser can construct the labeled representations.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"It differs from previous machine learning-based NERs in that it uses information from the whole document to classify each word, with just one classifier.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"These knowledge sources determine whether the contexts surrounding an anaphor and antecedent are compatible.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"97 81.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Thus in an English sentence such as I'm going to show up at the ACL one would reasonably conjecture that there are eight words separated by seven spaces.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Not all the layers have been produced for all the texts yet.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Inspection of the data shows that at n = 2500, the two classifiers both give labels on 44,281 (49.2%) of the unlabeled examples, and give the same label on 99.25% of these cases.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Unfortunately, there is no standard corpus of Chinese texts, tagged with either single or multiple human judgments, with which one can compare performance of various methods.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We follow the guidelines developed in the TIGER project (Brants et al. 2002) for syntactic annotation of German newspaper text, using the Annotate3 tool for interactive construction of tree structures.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The computing time, the number of search errors, and the multi-reference WER (mWER) are shown as a function of t0.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In a more recent study than Chang et al., Wang, Li, and Chang (1992) propose a surname-driven, non-stochastic, rule-based system for identifying personal names.17 Wang, Li, and Chang also compare their performance with Chang et al.'s system.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(a) ;IE shi4 'be' => ;IE;IE shi4bu2-shi4 (be-not-be) 'is it?'","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Evaluation metrics used are Attachment Score (AS), i.e. the proportion of tokens that are attached to the correct head, and Exact Match (EM), i.e. the proportion of sentences for which the dependency graph exactly matches the gold standard.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The following features were used: full-string=x The full string (e.g., for Maury Cooper, full- s tring=Maury_Cooper). contains(x) If the spelling contains more than one word, this feature applies for any words that the string contains (e.g., Maury Cooper contributes two such features, contains (Maury) and contains (Cooper) . allcapl This feature appears if the spelling is a single word which is all capitals (e.g., IBM would contribute this feature). allcap2 This feature appears if the spelling is a single word which is all capitals or full periods, and contains at least one period.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Section 4 draws some conclusions from the present state of the effort.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thus, as in CFG's, at any point in the derivation, the set of structures that can be applied is determined only by a finite set of rules encapsulated by the grammar.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Cohen and Smith approach this by introducing the \u03b1 hyperparameter, which performs best when optimized independently for each sentence (cf.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Note that Chang, Chen, and Chen (1991), in addition to word-frequency information, include a constraint-satisfication model, so their method is really a hybrid approach.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Learned Tag Prior (PRIOR) We next assume there exists a single prior distribution \u00cf\u0088 over tag assignments drawn from DIRICHLET(\u00ce\u00b2, K ).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The translation scores for the hypotheses generated with different threshold values t0 are compared to the translation scores obtained with a conservatively large threshold t0 = 10:0 . For each test series, we count the number of sentences whose score is worse than the corresponding score of the test series with the conservatively large threshold t0 = 10:0, and this number is reported as the number of search errors.","label":0},{"source_text":"Here we present two algorithms.","target_text":"For the moment we will assume that there are only two possible labels: each y, is in { \u20141, +1}.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For queries, we uniformly sampled 10 million hits and 10 million misses.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"level.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This smooth guarantees that there are no zeroes estimated.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"That is, given a choice between segmenting a sequence abc into abc and ab, c, the former will always be picked so long as its cost does not exceed the summed costs of ab and c: while; it is possible for abc to be so costly as to preclude the larger grouping, this will certainly not usually be the case.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We have not to date explored these various options.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Memory usage is likely much lower than ours. fThe original paper (Germann et al., 2009) provided only 2s of query timing and compared with SRI when it exceeded available RAM.","label":0},{"source_text":"There is no global pruning.","target_text":"The resulting algorithm is depicted in Table 1.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Recently, this topic has been getting more attention, as is evident from the Paraphrase Workshops in 2003 and 2004, driven by the needs of various NLP applications.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"As an indication, in our core corpus, we found an average sentence length of 15.8 words and 1.8 verbs per sentence, whereas a randomly taken sample of ten commentaries from the national papers Su\u00c2\u00a8ddeutsche Zeitung and Frankfurter Allgemeine has 19.6 words and 2.1 verbs per sentence.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"MENE has only been tested on MUC7.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"7 Conclusion and Future Work.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"For parsing, the most challenging form of ambiguity occurs at the discourse level.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Unexpectedly, because the ministries of treasury and education both had prepared the teacher plan together.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The question is how to normalize the probabilities in such a way that smaller groupings have a better shot at winning.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"This can be repeated several times to collect a list of author \/ book title pairs and expressions.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Under this scheme, n human judges are asked independently to segment a text.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"After the first step towards breadth had been taken with the PoS-tagging, RST annotation, and URML conversion of the entire corpus of 170 texts12 , emphasis shifted towards depth.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Consequently, we cut their evidence values in half to lessen their influence.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Our system fails in (a) because of$ shenl, a rare family name; the system identifies it as a family name, whereas it should be analyzed as part of the given name.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We have developed a general approach for combining parsers when preserving the entire structure of a parse tree is important.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"NN \u00ef\u00bf\u00bd .e NP NNP NP DTNNP NN \u00ef\u00bf\u00bd .e NP NP NNP NP Table 5: Evaluation of 100 randomly sampled variation nuclei types.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We map the ATB morphological analyses to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags for all experiments.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"\u00e2\u0080\u0094ested in correlations between prosody and dis course structure.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We compare the manually annotated grammar, which we incorporate into the Stanford parser, to both the Berkeley (Petrov et al., 2006) and Bikel (Bikel, 2004) parsers.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"It is likely that the more complex cases, where path information could make a difference, are beyond the reach of the parser in most cases.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Compared to the EMEA\/EP setting, the two domains in the NIST setting are less homogeneous and more similar to each other; there is also considerably more IN text available.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Word Head Of Complement POS 1 '01 inna \u00e2\u0080\u009cIndeed, truly\u00e2\u0080\u009d VP Noun VBP 2 '01 anna \u00e2\u0080\u009cThat\u00e2\u0080\u009d SBAR Noun IN 3 01 in \u00e2\u0080\u009cIf\u00e2\u0080\u009d SBAR Verb IN 4 01 an \u00e2\u0080\u009cto\u00e2\u0080\u009d SBAR Verb IN Table 1: Diacritized particles and pseudo-verbs that, after orthographic normalization, have the equivalent surface form 0 an.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Figure 3 shows a small fragment of the WFST encoding the dictionary, containing both entries forjust discussed, g:t\u00c2\u00a5 zhonglhua2 min2guo2 (China Republic) 'Republic of China,' and i\u00c2\u00a5inl.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Relative pronouns with only 1 NP in scope..","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"yields.6 But to its great advantage, it has a high ratio of non-terminals\/terminals (\u00ce\u00bc Constituents \/ \u00ce\u00bc Length).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In total, for the 2,000 NE category pairs, 5,184 keywords are found.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Since the work tapes store integers (which can be written in binary) that never exceed the size of the input, no configuration has space exceeding 0(log n).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We can make several observations on the cause of errors.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The local feature groups are: Non-Contextual Feature: This feature is set to 1 for all tokens.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Figure 3 shows examples of semantic expectations that were learned.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Although the ultimate goal of the Verbmobil project is the translation of spoken language, the input used for the translation experiments reported on in this paper is the (more or less) correct orthographic transcription of the spoken sentences.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The ability to redistribute belief values across sets rather than individual hypotheses is key.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"In the small window on the left, search queries can be entered, here one for an NP that has been annotated on the co-reference layer as bridging.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Comparison with state-of-the-art taggers For comparison we consider two unsupervised tag- gers: the HMM with log-linear features of Berg- Kirkpatrick et al.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If it starts with a lower case letter, and contains both upper and lower case letters, then (mixedCaps, zone) is set to 1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A totally non\u00c2\u00ad stochastic rule-based system such as Wang, Li, and Chang's will generally succeed in such cases, but of course runs the risk of overgeneration wherever the single-hanzi word is really intended.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In the first part of the experiment, dependency graphs from the treebanks were projectivized using the algorithm described in section 2.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"So far the graph has been completely unlabeled.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We have shown that, at least given independent human judgments, this is not the case, and that therefore such simplistic measures should be mistrusted.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In order to handle the necessary word reordering as an optimization problem within our dynamic programming approach, we describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming (Held, Karp, 1962).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We can only compare with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Quantization can be improved by jointly encoding probability and backoff.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In addition to the automatic methods, AG, GR, and ST, just discussed, we also added to the plot the values for the current algorithm using only dictionary entries (i.e., no productively derived words or names).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Compared to related work on the recovery of long-distance dependencies in constituency-based parsing, our approach is similar to that of Dienes and Dubey (2003) in that the processing of non-local dependencies is partly integrated in the parsing process, via an extension of the set of syntactic categories, whereas most other approaches rely on postprocessing only.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":".","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Motivated by these questions, we significantly raise baselines for three existing parsing models through better grammar engineering.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"To initialize the graph we tag the English side of the parallel text using a supervised model.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"SRILM inefficiently stores 64-bit pointers.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Table 8a shows that the best model recovers SBAR at only 71.0% F1.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This best instance-weighting model beats the equivalant model without instance weights by between 0.6 BLEU and 1.8 BLEU, and beats the log-linear baseline by a large margin.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"pronunciation depends upon word affiliation: tfJ is pronounced deO when it is a prenominal modification marker, but di4 in the word \u00c2\u00a7tfJ mu4di4 'goal'; fl; is normally ganl 'dry,' but qian2 in a person's given name.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"However, this argument is only plausible if the formal framework allows non-projective dependency structures, i.e. structures where a head and its dependents may correspond to a discontinuous constituent.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This has solutions: where pI(s|t) is derived from the IN corpus using relative-frequency estimates, and po(s|t) is an instance-weighted model derived from the OUT corpus.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"\u00ce\u00b8 has a belief value of 1.0, indicating complete certainty that the correct hypothesis is included in the set, and a plausibility value of 1.0, indicating that there is no evidence for competing hypotheses.5 As evidence is collected and the likely hypotheses are whittled down, belief is redistributed to subsets of \u00ce\u00b8.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For even larger models, we recommend RandLM; the memory consumption of the cache is not expected to grow with model size, and it has been reported to scale well.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"To check for this, we do pairwise bootstrap resampling: Again, we repeatedly sample sets of sentences, this time from both systems, and compare their BLEU scores on these sets.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For definite NPs, the results are a mixed bag: some knowledge sources increased recall a little, but at the expense of some precision.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"However, there is a strong relationship between ni1s and the number of hanzi in the class.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The cities of the traveling salesman problem correspond to source Table 1: DP algorithm for statistical machine translation.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Thus we are interested not in extraction, but actual generation from representations that may be developed to different degrees of granularity.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"One hybridization strategy is to let the parsers vote on constituents' membership in the hypothesized set.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Beyond optimizing the memory size of TRIE, there are alternative data structures such as those in Guthrie and Hepple (2010).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"This work has been supported as part of the Verbmobil project (contract number 01 IV 601 A) by the German Federal Ministry of Education, Science, Research and Technology and as part of the Eutrans project (ESPRIT project number 30268) by the European Community.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Each model was able to produce hypotheses for all input sentences.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"We have shown that the maximum entropy framework is able to use global information directly.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank, leaving only sections 22 and 23 completely untouched during the development of any of the parsers.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"A crucial difference is that the number of parameters is greatly reduced as is the number of variables that are sampled during each iteration.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Each extraction pattern represents a linguistic expression and a syntactic position indicating where a role filler can be found.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Unfortunately, there is no standard corpus of Chinese texts, tagged with either single or multiple human judgments, with which one can compare performance of various methods.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"For example, out of 905 phrases in the CC- domain, 211 phrases contain keywords found in step 2.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This Good\u00c2\u00ad Turing estimate of p(unseen(f,) If,) can then be used in the normal way to define the probability of finding a novel instance of a construction in ir, in a text: p(unseen(f,)) = p(unseen(f,) I f,) p(fn Here p(ir,) is just the probability of any construction in ft as estimated from the frequency of such constructions in the corpus.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This is not ideal for some applications, however.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"3.1 Maximum Entropy.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"On the other hand, in a translation system one probably wants to treat this string as a single dictionary word since it has a conventional and somewhat unpredictable translation into English.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We train and test on the CoNLL-X training set.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"including Third Tone Sandhi (Shih 1986), which changes a 3 (low) tone into a 2 (rising) tone before another 3 tone: 'j\";gil, xiao3 [lao3 shu3] 'little rat,' becomes xiao3 { lao2shu3 ], rather than xiao2 { lao2shu3 ], because the rule first applies within the word lao3shu3 'rat,' blocking its phrasal application.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Linear weights are difficult to incorporate into the standard MERT procedure because they are \u201chidden\u201d within a top-level probability that represents the linear combination.1 Following previous work (Foster and Kuhn, 2007), we circumvent this problem by choosing weights to optimize corpus loglikelihood, which is roughly speaking the training criterion used by the LM and TM themselves.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The relativizer f(\u201cthat\u201d) for example, may attach to an arbitrarily long relative clause that goes beyond token boundaries.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Firstly, Hebrew unknown tokens are doubly unknown: each unknown token may correspond to several segmentation possibilities, and each segment in such sequences may be able to admit multiple PoS tags.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We maintain a separate array for each length n containing all n-gram entries sorted in suffix order.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"It is a sequence of proper nouns within an NP; its last word Cooper is the head of the NP; and the NP has an appositive modifier (a vice president at S.&P.) whose head is a singular noun (president).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"This design does not guarantee \u00e2\u0080\u009cstructural zeros,\u00e2\u0080\u009d but biases towards sparsity.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(Blum and Mitchell 98) give an example that illustrates just how powerful the second constraint can be.","label":0},{"source_text":"There is no global pruning.","target_text":"When translating the sentence monotonically from left to right, the translation of the German finite verb 'kann', which is the left verbal brace in this case, is postponed until the German noun phrase 'mein Kollege' is translated, which is the subject of the sentence.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Crucially, the conventional orthographic form of MSA text is unvocalized, a property that results in a deficient graphical representation.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Finally, a DempsterShafer probabilistic model evaluates the evidence provided by the knowledge sources for all candidate antecedents and makes the final resolution decision.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Table 6: Incremental dev set results for the manually annotated grammar (sentences of length \u00e2\u0089\u00a4 70).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"+ cost(unseen(fm, as desired.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Second, we identified the 100 most frequent nouns in the training corpus and manually labeled them with semantic tags.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Unigram lookup is dense so we use an array of probability and backoff values.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The equation for sampling a single type-level assignment Ti is given by, 0.2 0 5 10 15 20 25 30 Iteration Figure 2: Graph of the one-to-one accuracy of our full model (+FEATS) under the best hyperparameter setting by iteration (see Section 5).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"86 78.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"This group consists of (3 total number of possible zones) features.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This is consistent with the nature of these two settings: log-linear combination, which effectively takes the intersection of IN and OUT, does relatively better on NIST, where the domains are broader and closer together.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"While we cannot prove there are no such useful features on which one should condition trust, we can give some insight into why the features we explored offered no gain.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"18 We are grateful to ChaoHuang Chang for providing us with this set.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"(2009b) evaluated the Bikel parser using the same ATB split, but only reported dev set results with gold POS tags for sentences of length \u00e2\u0089\u00a4 40.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"needs to be in initCaps to be considered for this feature.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In (b) is a plausible segmentation for this sentence; in (c) is an implausible segmentation.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The final model tions.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We hypothesize that modeling morphological information will greatly constrain the set of possible tags, thereby further refining the representation of the tag lexicon.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Both BBN and NYU have tagged their own data to supplement the official training data.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In this section, we describe the seven general knowledge sources and explain how the DempsterShafer model makes resolutions.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"How should the absence of vowels and syntactic markers influence annotation choices and grammar development?","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We again assume a training set of n examples {x1 . xri} where the first m examples have labels {y1 ... yin}, and the last (n \u2014 m) examples are unlabeled.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The second term is a regularizer and encourages all type marginals to be uniform to the extent that is allowed by the first two terms (cf. maximum entropy principle).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"(1), Pr(eI 1) is the language model, which is a trigram language model in this case.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"f, nan2gual+men0 'pumpkins' is by no means impossible.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"we perform five runs with different random initialization of sampling state.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Table 3: Training and test conditions for the Verbmobil task (*number of words without punctuation marks).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"5 60.6 Table 3: Multilingual Results: We report token-level one-to-one and many-to-one accuracy on a variety of languages under several experimental settings (Section 5).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The gender, number, and scoping KSs eliminate candidates from consideration.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"5 70.1 58.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"96 75.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"(c) After they blindfolded the men...","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The scoping heuristics are based on the anaphor type: for reflexive pronouns the scope is the current clause, for relative pronouns it is the prior clause following its VP, for personal pronouns it is the anaphor\u00e2\u0080\u0099s sentence and two preceding sentences, and for definite NPs it is the anaphor\u00e2\u0080\u0099s sentence and eight preceding sentences.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The results of this experiment can be seen in Table 5.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The supervised POS tagging accuracies (on this tagset) are shown in the last row of Table 2.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Different annotations of the same text are mapped into the same data structure, so that search queries can be formulated across annotation levels.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The 2(Yarowsky 95) describes the use of more sophisticated smoothing methods.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The global feature groups are: InitCaps of Other Occurrences (ICOC): There are 2 features in this group, checking for whether the first occurrence of the same word in an unambiguous position (non first-words in the TXT or TEXT zones) in the same document is initCaps or not-initCaps.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"2.2.2 The Caseframe Network The first type of contextual role knowledge that BABAR learns is the Caseframe Network (CFNet), which identifies caseframes that co-occur in anaphor\/antecedent resolutions.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"of Articles No.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Values in the trie are minimally sized at the bit level, improving memory consumption over trie implementations in SRILM, IRSTLM, and BerkeleyLM.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For even larger models, storing counts (Talbot and Osborne, 2007; Pauls and Klein, 2011; Guthrie and Hepple, 2010) is a possibility.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We are unaware of prior results for the Stanford parser.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"68 95.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"With a human evaluation we also showed that ATB inter-annotator agreement remains low relative to the WSJ corpus.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The label for a test example with features x is then defined as In this paper we define h(x, y) as the following function of counts seen in training data: Count(x,y) is the number of times feature x is seen with label y in training data, Count(x) = EyEy Count(x, y). a is a smoothing parameter, and k is the number of possible labels.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Special thanks to Jan Haji\u02c7c and Matthias Trautner Kromann for assistance with the Czech and Danish data, respectively, and to Jan Haji\u02c7c, Tom\u00b4a\u02c7s Holan, Dan Zeman and three anonymous reviewers for valuable comments on a preliminary version of the paper.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"A corpus of German newspaper commentaries has been assembled and annotated with different information (and currently, to different degrees): part-of-speech, syntax, rhetorical structure, connectives, co-reference, and information structure.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"68 95.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"To bridge this gap, we consider a practically motivated scenario, in which we want to leverage existing resources from a resource-rich language (like English) when building tools for resource-poor foreign languages.1 We assume that absolutely no labeled training data is available for the foreign language of interest, but that we have access to parallel data with a resource-rich language.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"In an experiment on automatic rhetorical parsing, the RST-annotations and PoS tags were used by (Reitter 2003) as a training corpus for statistical classification with Support Vector Machines.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We are especially grateful to Taylor Berg- Kirkpatrick for running additional experiments.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The idea is to have a pipeline of shallow-analysis modules (tagging, chunk- ing, discourse parsing based on connectives) and map the resulting underspecified rhetorical tree (see Section 2.4) into a knowledge base that may contain domain and world knowledge for enriching the representation, e.g., to resolve references that cannot be handled by shallow methods, or to hypothesize coherence relations.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Church and Hanks [1989]), and we have used lists of character pairs ranked by mutual information to expand our own dictionary.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"92 76.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In most cases, however, these expansions come with a steep increase in model complexity, with respect to training procedure and inference time.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We use N(u) to denote the neighborhood of vertex u, and fixed n = 5 in our experiments.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"0750271 and by the DARPA GALE program.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We retain segmentation markers\u00e2\u0080\u0094which are consistent only in the vocalized section of the treebank\u00e2\u0080\u0094to differentiate between e.g. \u00ef\u00bf\u00bd \u00e2\u0080\u009cthey\u00e2\u0080\u009d and \u00ef\u00bf\u00bd + \u00e2\u0080\u009ctheir.\u00e2\u0080\u009d Because we use the vocalized section, we must remove null pronoun markers.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"95 76.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"For this experiment, we compare our model with the uniform tag assignment prior (1TW) with the learned prior (+PRIOR).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"This formulation of the constraint feature is equivalent to the use of a tagging dictionary extracted from the graph using a threshold T on the posterior distribution of tags for a given word type (Eq.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"We developed a first version of annotation guidelines for co-reference in PCC (Gross 2003), which served as basis for annotating the core corpus but have not been empirically evaluated for inter-annotator agreement yet.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Finally, we make some improvements to baseline approaches.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We first give a brief overview of boosting algorithms.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We would also like to thank Amarnag Subramanya for helping us with the implementation of label propagation and Shankar Kumar for access to the parallel data.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Acknowledgments We thank Meni Adler and Michael Elhadad (BGU) for helpful comments and discussion.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Section 3 describes the complete coreference resolution model, which uses the contextual role knowledge as well as more traditional coreference features.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Note that the backoff model assumes that there is a positive correlation between the frequency of a singular noun and its plural.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The results are given in Table 4.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Depending on the threshold t0, the search algorithm may miss the globally optimal path which typically results in additional translation errors.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"MENE without Proteus, however, did not do very well and only achieved an F measure of 84.22% (Borthwick, 1999).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Specifically, we (+FEATS) utilizes the tag prior as well as features (e.g., suffixes and orthographic features), discussed in Section 3, for the P (W |T , \u00cf\u0088) component.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We present several variations for the lexical component P (T , W |\u00cf\u0088), each adding more complex pa rameterizations.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"While our method also enforces a singe tag per word constraint, it leverages the transition distribution encoded in an HMM, thereby benefiting from a richer representation of context.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The effect of a second reference resolution classifier is not entirely the same as that of global features.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The method uses a &quot;soft&quot; measure of the agreement between two classifiers as an objective function; we described an algorithm which directly optimizes this function.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Although the tag distributions of the foreign words (Eq.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Sima\u2019an et al. (2001) presented parsing results for a DOP tree-gram model using a small data set (500 sentences) and semiautomatic morphological disambiguation.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The first unsupervised algorithm we describe is based on the decision list method from (Yarowsky 95).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"A modified language model probability p\u00c3\u0086(eje0; e00) is defined as follows: p\u00c3\u0086(eje0; e00) =  1:0 if \u00c3\u0086 = 0 p(eje0; e00) if \u00c3\u0086 = 1 : We associate a distribution p(\u00c3\u0086) with the two cases \u00c3\u0086 = 0 and \u00c3\u0086 = 1 and set p(\u00c3\u0086 = 1) = 0:7.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"It is chosen such that the decisions it made in including or excluding constituents are most probable under the models for all of the parsers.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Consequently, all three parsers prefer the nominal reading.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Table 3 Classes of words found by ST for the test corpus.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"A large number of rules is needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classi- However, we show that the use of data can reduce the requirements for supervision to just 7 simple &quot;seed&quot; rules.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The ATB annotation guidelines specify that proper nouns should be specified with a flat NP (a).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Here, the term frequency (TF) is the frequency of a word in the bag and the inverse term frequency (ITF) is the inverse of the log of the frequency in the entire corpus.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"This departure from the traditional token-based tagging approach allows us to explicitly capture type- level distributional properties of valid POS tag as signments as part of the model.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(3) shows learning curves for CoBoost.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"We use MMAX for this annotation as well.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"9 65.5 46.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"yu2 'fish.'","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, it is almost universally the case that no clear definition of what constitutes a \"correct\" segmentation is given, so these performance measures are hard to evaluate.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Recent work by Finkel and Manning (2009) which re-casts Daum\u00b4e\u2019s approach in a hierarchical MAP framework may be applicable to this problem.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Since these are distinct syntactic units, they are typically segmented.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Given a sorted array A, these other packages use binary search to find keys in O(log |A|) time.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Finally, the statistical method fails to correctly group hanzi in cases where the individual hanzi comprising the name are listed in the dictionary as being relatively high-frequency single-hanzi words.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Here, the pruning threshold t0 = 10:0 is used.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Mikheev et al.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"2.6 Co-reference.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Because we are interested in applying our techniques to languages for which no labeled resources are available, we paid particular attention to minimize the number of free parameters and used the same hyperparameters for all language pairs.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Because of their size, the examples (Figures 2 to 4) appear at the end of the paper.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"9 www.ling.unipotsdam.de\/sfb\/ Figure 2: Screenshot of Annis Linguistic Database 3.3 Symbolic and knowledge-based.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In the labeled version of these metrics (L) both heads and arc labels must be correct, while the unlabeled version (U) only considers heads.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The semilinearity of Tree Adjoining Languages (TAL's), MCTAL's, and Head Languages (HL's) can be proved using this property, with suitable restrictions on the composition operations.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Among these are words derived by various productive processes, including: 1.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In contrast, NNP (proper nouns) form a large portion of vocabulary.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We refer to different readings as different analyses whereby the segments are deterministic given the sequence of PoS tags.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Contextual role knowledge provides evidence as to whether a candidate is a plausible antecedent for an anaphor.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"To our knowledge, ours is the first analysis of this kind for Arabic parsing.","label":1},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We map the ATB morphological analyses to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags for all experiments.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"AdaBoost is given access to a weak learning algorithm, which accepts as input the training examples, along with a distribution over the instances.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The prediction based on these features is a knearest neighbor classification, using the IB1 algorithm and k = 5, the modified value difference metric (MVDM) and class voting with inverse distance weighting, as implemented in the TiMBL software package (Daelemans et al., 2003).","label":0},{"source_text":"There is no global pruning.","target_text":"The resulting algorithm has a complexity of O(n!).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We used 22 features for the logistic weighting model, divided into two groups: one intended to reflect the degree to which a phrase pair belongs to general language, and one intended to capture similarity to the IN domain.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Reference resolution involves finding words that co-refer to the same entity.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The difference in performance between pronouns and definite noun phrases surprised us.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"They also describe an application of cotraining to classifying web pages (the to feature sets are the words on the page, and other pages pointing to the page).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"It then computes a normalized Levenshtein edit distance between the extracted chain and the reference.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Sometime, multiple words are needed, like \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, \u00e2\u0080\u009cprime minister\u00e2\u0080\u009d or \u00e2\u0080\u009cpay for\u00e2\u0080\u009d (\u00e2\u0080\u009cpay\u00e2\u0080\u009d and \u00e2\u0080\u009cpay for\u00e2\u0080\u009d are different senses in the CC-domain).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Both BBN and NYU have tagged their own data to supplement the official training data.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"AdaBoost.MH can be applied to the problem using these pseudolabels in place of supervised examples.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Recent research has demonstrated that incorporating this sparsity constraint improves tagging accuracy.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Clearly, retaining the original frequencies is important for good performance, and globally smoothing the final weighted frequencies is crucial.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Email: rlls@bell-labs.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"For all languages, the vocabulary sizes increase by several thousand words.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Email: rlls@bell-labs.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Each out-of-domain phrase pair is characterized by a set of simple features intended to reflect how useful it will be.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"To investigate the influence of these factors, we analyze Modern Standard Arabic (henceforth MSA, or simply \u00e2\u0080\u009cArabic\u00e2\u0080\u009d) because of the unusual opportunity it presents for comparison to English parsing results.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Again, famous place names will most likely be found in the dictionary, but less well-known names, such as 1PM\u00c2\u00b1 R; bu4lang3-shi4wei2-ke4 'Brunswick' (as in the New Jersey town name 'New Brunswick') will not generally be found.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Finally, this effort is part of a much larger program that we are undertaking to develop stochastic finite-state methods for text analysis with applications to TIS and other areas; in the final section of this paper we will briefly discuss this larger program so as to situate the work discussed here in a broader context.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Following the system devised under the Qing emperor Kang Xi, hanzi have traditionally been classified according to a set of approximately 200 semantic radicals; members of a radical class share a particular structural component, and often also share a common meaning (hence the term 'semantic').","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The combining technique must act as a multi-position switch indicating which parser should be trusted for the particular sentence.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We make the assumption that for each example, both xi,. and x2,2 alone are sufficient to determine the label yi.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In order to pass these constraints onto the parser, the lexical rules in the grammar are of the form pi \u2014* (si, pi) Parameter Estimation The grammar probabilities are estimated from the corpus using simple relative frequency estimates.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"27 80.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In Section 5, we try to compare results of MENE, IdentiFinder, and MENERGI.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Next, for each pair of NE categories, we collect all the contexts and find the keywords which are topical for that NE category pair.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, BABAR learned that agents that \u00e2\u0080\u009cassassinate\u00e2\u0080\u009d or \u00e2\u0080\u009cinvestigate a cause\u00e2\u0080\u009d are usually humans or groups (i.e., organizations).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In MUC6 and MUC7, the named entity task is defined as finding the following classes of names: person, organization, location, date, time, money, and percent (Chinchor, 1998; Sundheim, 1995) Machine learning systems in MUC6 and MUC 7 achieved accuracy comparable to rule-based systems on the named entity task.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"In addition to the optimizations specific to each datastructure described in Section 2, we implement several general optimizations for language modeling.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Personal names such as 00, 3R; zhoulenl-lai2 'Zhou Enlai.'","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"If the parse contains productions from outside our grammar the machine has no direct method for handling them (e.g. the resulting database query may be syntactically malformed).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"IRSTLM 5.60.02 (Federico et al., 2008) is a sorted trie implementation designed for lower memory consumption.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"This group contains a large number of features (one for each token string present in the training data).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In Section 6 we dis\u00c2\u00ad cuss other issues relating to how higher-order language models could be incorporated into the model.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In Input: {(x1,i, Initialize: Vi, j : e(xi) = 0.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The estimation of the probabilities in the model is carried out as shown in Equation 4.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Since MUC6, BBN' s Hidden Markov Model (HMM) based IdentiFinder (Bikel et al., 1997) has achieved remarkably good performance.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We outlined the definition of a family of constrained grammatical formalisms, called Linear Context-Free Rewriting Systems.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Morphologically derived words such as, xue2shengl+men0.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We then gather all phrases with the same keyword.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Future work along these lines will incorporate other layers of annotation, in particular the syntax information.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Benchmarks use the package\u2019s binary format; our code is also the fastest at building a binary file.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Queries detect the invalid probability, using the node only if it leads to a longer match.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Their work used subject-verb, verb-object, and adjective-noun relations to compare the contexts surrounding an anaphor and candidate.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"32 81.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Here, we process only full-form words within the translation procedure.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"More recently, (Riloff and Jones 99) describe a method they term &quot;mutual bootstrapping&quot; for simultaneously constructing a lexicon and contextual extraction patterns.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Unbounded dependencies between branches are not possible in such a system.","label":0},{"source_text":"Here we present two algorithms.","target_text":"More recently, (Riloff and Jones 99) describe a method they term &quot;mutual bootstrapping&quot; for simultaneously constructing a lexicon and contextual extraction patterns.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The following context-free production captures the derivation step of the grammar shown in Figure 7, in which the trees in the auxiliary tree set are adjoined into themselves at the root node (address c).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Figure 2 shows examples of lexical expectations that were learned for both domains.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Words and punctuation that appear in brackets are considered optional.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We experiment with four values for each hyperparameter resulting in 16 (\u00ce\u00b1, \u00ce\u00b2) combinations: \u00ce\u00b1 \u00ce\u00b2 0.001, 0.01, 0.1, 1.0 0.01, 0.1, 1.0, 10 Iterations In each run, we performed 30 iterations of Gibbs sampling for the type assignment variables W .4 We use the final sample for evaluation.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Specifically, for both settings we report results on the median run for each setting.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Table 3 shows BABAR\u00e2\u0080\u0099s performance when the four contextual role knowledge sources are added.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"For these models we limit the options provided for OOV words by not considering the entire token as a valid segmentation in case at least some prefix segmentation exists.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This latter evaluation compares the performance of the system with that of several human judges since, as we shall show, even people do not agree on a single correct way to segment a text.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We model po(s|t) using a MAP criterion over weighted phrase-pair counts: and from the similarity to (5), assuming y = 0, we see that w\u03bb(s, t) can be interpreted as approximating pf(s, t)\/po(s, t).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The number of latent HMM states for each language in our experiments was set to the number of fine tags in the language\u2019s treebank.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Training under this model involves estimation of parameter values for P(y), P(m) and P(x I y).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"If enough parsers suggest that a particular constituent belongs in the parse, we include it.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"AdaBoost finds a weighted combination of simple (weak) classifiers, where the weights are chosen to minimize a function that bounds the classification error on a set of training examples.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"To optimize this function, we used L-BFGS, a quasi-Newton method (Liu and Nocedal, 1989).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"hanzi in the various name positions, derived from a million names.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In a fully supervised setting, the task is to learn a function f such that for all i = 1...m, f (xi,i, 12,i) = yz.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE has only been tested on MUC7.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"(2010), we adopt a simpler na\u00c2\u00a8\u00c4\u00b1ve Bayes strategy, where all features are emitted independently.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This is a unique object for which we are able to define a proper probability model.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Contextual role knowledge provides evidence as to whether a candidate is a plausible antecedent for an anaphor.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We also report word type level accuracy, the fraction of word types assigned their majority tag (where the mapping between model state and tag is determined by greedy one-to-one mapping discussed above).5 For each language, we aggregate results in the following way: First, for each hyperparameter setting, evaluate three variants: The first model (1TW) only 4 Typically, the performance stabilizes after only 10 itera-.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Section 4 draws some conclusions from the present state of the effort.","label":0},{"source_text":"There is no global pruning.","target_text":"A procedural definition to restrict1In the approach described in (Berger et al., 1996), a mor phological analysis is carried out and word morphemes rather than full-form words are used during the search.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We include a constituent in our hypothesized parse if it appears in the output of a majority of the parsers.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A morpheme, on the other hand, usually corresponds to a unique hanzi, though there are a few cases where variant forms are found.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Gan's solution depends upon a fairly sophisticated language model that attempts to find valid syntactic, semantic, and lexical relations between objects of various linguistic types (hanzi, words, phrases).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"30 16.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"More recently, (Riloff and Jones 99) describe a method they term &quot;mutual bootstrapping&quot; for simultaneously constructing a lexicon and contextual extraction patterns.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In figure 4, reverse relations are indicated by `*\u00e2\u0080\u0099 next to the frequency.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"3.3 Evaluation Results.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This extends previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and using a simpler training procedure.","label":0},{"source_text":"There is no global pruning.","target_text":"The Verbmobil task is an appointment scheduling task.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We developed a first version of annotation guidelines for co-reference in PCC (Gross 2003), which served as basis for annotating the core corpus but have not been empirically evaluated for inter-annotator agreement yet.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The difference in precision between similarity and Bayes switching techniques is significant, but the difference in recall is not.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, if X and Y are coreferent, then both X and Y are considered to co-occur with the caseframe that extracts X as well as the caseframe that extracts Y. We will refer to the set of nouns that co-occur with a caseframe as the lexical expectations of the case- frame.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"These models generally outperform our memory consumption but are much slower, even when cached.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We introduce several new ideas.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We respond to this on the one hand with a format for its underspecification (see 2.4) and on the other hand with an additional level of annotation that attends only to connectives and their scopes (see 2.5), which is intended as an intermediate step on the long road towards a systematic and objective treatment of rhetorical structure.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Their work used subject-verb, verb-object, and adjective-noun relations to compare the contexts surrounding an anaphor and candidate.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"0 70.9 42.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"First, we parsed the training corpus, collected all the noun phrases, and looked up each head noun in WordNet (Miller, 1990).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Precision is the portion of hypothesized constituents that are correct and recall is the portion of the Treebank constituents that are hypothesized.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"To combat the proliferation of parsing edges, we prune the lattices according to a hand-constructed lexicon of 31 clitics listed in the ATB annotation guidelines (Maamouri et al., 2009a).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":".","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We adopted the MUC6 guidelines for evaluating coreference relationships based on transitivity in anaphoric chains.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"For example, one parser could be more accurate at predicting noun phrases than the other parsers.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"For brevity we omit the segments from the analysis, and so analysis of the form \u201cfmnh\u201d as f\/REL mnh\/VB is represented simply as REL VB.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"0 55.3 34.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"For example, in Information Retrieval (IR), we have to match a user\u00e2\u0080\u0099s query to the expressions in the desired documents, while in Question Answering (QA), we have to find the answer to the user\u00e2\u0080\u0099s question even if the formulation of the answer in the document is different from the question.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"2.1.1 Lexical Seeding It is generally not safe to assume that multiple occurrences of a noun phrase refer to the same entity.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"An alternate approximation to (8) would be to let w,\\(s, t) directly approximate p\u02c6I(s, t).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The feature-based model replaces the emission distribution with a log-linear model, such that: on the word identity x, features checking whether x contains digits or hyphens, whether the first letter of x is upper case, and suffix features up to length 3.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In this case, Maury Cooper is extracted.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Models To assess the marginal utility of each component of the model (see Section 3), we incremen- tally increase its sophistication.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Subsets C of increasing cardinality c are processed.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Full Chinese personal names are in one respect simple: they are always of the form family+given.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"More formally, we start by representing the dictionary D as a Weighted Finite State Trans\u00c2\u00ad ducer (WFST) (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"A more forceful approach for encoding sparsity is posterior regularization, which constrains the posterior to have a small number of expected tag assignments (Grac\u00c2\u00b8a et al., 2009).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"selected; and that recall is defined to be the number of correct hits divided by the number of items that should have been selected.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"1 1 0.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We call this approach parse hybridization.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"It would have therefore also been possible to use the integer programming (IP) based approach of Ravi and Knight (2009) instead of the feature-HMM for POS induction on the foreign side.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We loosely describe the class of all such systems as Linear Context-Free Rewriting Formalisms.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"For speed, we plan to implement the direct-mapped cache from BerkeleyLM.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"All four of the techniques studied result in parsing systems that perform better than any previously reported.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Our derivation is slightly different from the one presented in (Schapire and Singer 98) as we restrict at to be positive.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Given a sorted array A, these other packages use binary search to find keys in O(log |A|) time.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Topicalization of NP subjects in SVO configurations causes confusion with VO (pro-drop).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"36.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"f1; ; Jg denotes a coverage set including all positions from the starting position 1 to position J and j 2 fJ \u00f4\u0080\u0080\u0080L; ; Jg.","label":0},{"source_text":"There is no global pruning.","target_text":"the number of permutations carried out for the word reordering is given.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Particular relations are also consistent with particular hypotheses about the segmentation of a given sentence, and the scores for particular relations can be incremented or decremented depending upon whether the segmentations with which they are consistent are \"popular\" or not.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"[Hasegawa et al. 04] reported only on relation discovery, but one could easily acquire para phrases from the results.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The resulting algorithm has a complexity of O(n!).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"KS Function Ge nde r filters candidate if gender doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Suffixes and Prefixes: This group contains only two features: Corporate-Suffix and Person-Prefix.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We evaluate our model on seven languages exhibiting substantial syntactic variation.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Mixing, smoothing, and instance-feature weights are learned at the same time using an efficient maximum-likelihood procedure that relies on only a small in-domain development corpus.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":".","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Richer tag sets have been suggested for modeling morphologically complex distinctions (Diab, 2007), but we find that linguistically rich tag sets do not help parsing.","label":0},{"source_text":"There is no global pruning.","target_text":"For the error counts, a range from 0:0 to 1:0 is used.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"At each point during the derivation, the parser has a choice between pushing the next input token onto the stack \u2013 with or without adding an arc from the token on top of the stack to the token pushed \u2013 and popping a token from the stack \u2013 with or without adding an arc from the next input token to the token popped.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The hyperparameters \u00ce\u00b1 and \u00ce\u00b2 represent the concentration parameters of the token- and type-level components of the model respectively.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Figure 1: Translation of PCC sample commentary (STTS)2.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We have modified Moses (Koehn et al., 2007) to keep our state with hypotheses; to conserve memory, phrases do not keep state.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"At each point during the derivation, the prediction is based on six word tokens, the two topmost tokens on the stack, and the next four input tokens.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In Eq.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"However, such an approach requires a number of hand-coded systems, which may not be available in languages other than English.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For a word whose initCaps might be due to its position rather than its meaning (in headlines, first word of a sentence, etc), the case information of other occurrences might be more accurate than its own.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Learned Tag Prior (PRIOR) We next assume there exists a single prior distribution \u00cf\u0088 over tag assignments drawn from DIRICHLET(\u00ce\u00b2, K ).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"markBaseNP indicates these non-recursive nominal phrases.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"But we also need an estimate of the probability for a non-occurring though possible plural form like i\u00c2\u00a5JJ1l.f, nan2gua1-men0 'pumpkins.'","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"And indeed, converging on annotation guidelines is even more difficult than it is with co-reference.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The linear LM (lin lm), TM (lin tm) and MAP TM (map tm) used with non-adapted counterparts perform in all cases slightly worse than the log-linear combination, which adapts both LM and TM components.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"As seen by the drop in average individual parser performance baseline, the introduced parser does not perform very well.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"In our case multi-threading is trivial because our data structures are read-only and uncached.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Table 4 Differences in performance between our system and Wang, Li, and Chang (1992).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Morphologically derived words such as, xue2shengl+men0.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"a classifier.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"As noted in Section 1, our code finds the longest matching entry wnf for query p(wn|s(wn\u22121 f ) The probability p(wn|wn\u22121 f ) is stored with wnf and the backoffs are immediately accessible in the provided state s(wn\u22121 When our code walks the data structure to find wnf , it visits wnn, wnn\u22121, ... , wnf .","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, they list two sets, one consisting of 28 fragments and the other of 22 fragments, in which they had 0% recall and precision.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This is less effective in our setting, where IN and OUT are disparate.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Besides size of training data, the use of dictionaries is another factor that might affect performance.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"As expected, the most informative encoding, Head+Path, gives the highest accuracy with over 99% of all non-projective arcs being recovered correctly in both data sets.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"both MENE and IdentiFinder used more training data than we did (we used only the official MUC 6 and MUC7 training data).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"x EG, has agreed to be bought by H x EG, now owned by H x H to acquire EG x H\u00e2\u0080\u0099s agreement to buy EG Three of those phrases are actually paraphrases, but sometime there could be some noise; such as the second phrase above.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"This measure has the advantage of being completely automatic.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The following context-free production captures the derivation step of the grammar shown in Figure 7, in which the trees in the auxiliary tree set are adjoined into themselves at the root node (address c).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The frequency of the Company \u00e2\u0080\u0093 Company domain ranks 11th with 35,567 examples.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In such cases we assign all of the estimated probability mass to the form with the most likely pronunciation (determined by inspection), and assign a very small probability (a very high cost, arbitrarily chosen to be 40) to all other variants.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"For each 13 Of course, this weighting makes the PCFG an improper distribution.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"In contrast to these approaches, our method directly incorporates these constraints into the structure of the model.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The function A : F \u2014* C maps from the language specific fine-grained tagset F to the coarser universal tagset C and is described in detail in \u00a76.2: Note that when tx(y) = 1 the feature value is 0 and has no effect on the model, while its value is \u2212oc when tx(y) = 0 and constrains the HMM\u2019s state space.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Prague Dependency Treebank (Haji\u02c7c et al., 2001b), Danish Dependency Treebank (Kromann, 2003), and the METU Treebank of Turkish (Oflazer et al., 2003), which generally allow annotations with nonprojective dependency structures.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The model is often further restricted so that each source word is assigned to exactly one target word (Brown et al., 1993; Ney et al., 2000).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"123 examples fell into the noise category.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The Potsdam Commentary Corpus","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The translation search is carried out with the category markers and the city names are resubstituted into the target sentence as a postprocessing step.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"English parsing evaluations usually report results on sentences up to length 40.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"While Gan's system incorporates fairly sophisticated models of various linguistic information, it has the drawback that it has only been tested with a very small lexicon (a few hundred words) and on a very small test set (thirty sentences); there is therefore serious concern as to whether the methods that he discusses are scalable.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We are currently exploring such algorithms.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"If is seen infrequently during training (less than a small count), then will not be selected as a feature and all features in this group are set to 0.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The first concerns how to deal with ambiguities in segmentation.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Table 5: Effect of the beam threshold on the number of search errors (147 sentences).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"A totally non\u00c2\u00ad stochastic rule-based system such as Wang, Li, and Chang's will generally succeed in such cases, but of course runs the risk of overgeneration wherever the single-hanzi word is really intended.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The idea of distinguishing between general and domain-specific examples is due to Daum\u00b4e and Marcu (2006), who used a maximum-entropy model with latent variables to capture the degree of specificity.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"\u00c2\u00a3 : _ADV: 5.88 If:!","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"While the proportion of sentences containing non-projective dependencies is often 15\u201325%, the total proportion of non-projective arcs is normally only 1\u20132%.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"gaolbu4-gaolxing4 (hap-not-happy) 'happy?'","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Lexicon Feature: The string of the token is used as a feature.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"We apply a beam search concept as in speech recognition.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"One of the strengths of the DempsterShafer model is its natural ability to recognize when several credible hypotheses are still in play.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Second, the reduced number of hidden variables and parameters dramatically speeds up learning and inference.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"This feature has a linguistic justification.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"7 Unlike Dickinson (2005), we strip traces and only con-.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The difference between the featureless model (+PRIOR) and our full model (+FEATS) is 13.6% and 7.7% average error reduction on best and median settings respectively.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"For each language under consideration, Petrov et al. (2011) provide a mapping A from the fine-grained language specific POS tags in the foreign treebank to the universal POS tags.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The first step in the learning process is to generate training examples consisting of anaphor\/antecedent resolutions.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"2.3 Rhetorical structure.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Table 2 shows these similarity measures.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"2.6 Co-reference.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"f;g denotes the empty set, where no source sentence position is covered.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Among machine learning-based NERs, Identi- Finder has proven to be the best on the official MUC6 and MUC7 test data.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"RandLM is the clear winner in RAM utilization, but is also slower and lower quality.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"There are clearly eight orthographic words in the example given, but if one were doing syntactic analysis one would probably want to consider I'm to consist of two syntactic words, namely I and am.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"As each global feature group is added to the list of features, we see improvements to both MUC6 and MUC6 MUC7 Baseline 90.75% 85.22% + ICOC 91.50% 86.24% + CSPP 92.89% 86.96% + ACRO 93.04% 86.99% + SOIC 93.25% 87.22% + UNIQ 93.27% 87.24% Table 3: F-measure after successive addition of each global feature group Table 5: Comparison of results for MUC6 Systems MUC6 MUC7 No.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"information structure.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Firstly, Hebrew unknown tokens are doubly unknown: each unknown token may correspond to several segmentation possibilities, and each segment in such sequences may be able to admit multiple PoS tags.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For irt the Good-Turing estimate just discussed gives us an estimate of p(unseen(f,) I f,)-the probability of observing a previously unseen instance of a construction in ft given that we know that we have a construction in f,.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The foreign language vertices (denoted by Vf) correspond to foreign trigram types, exactly as in Subramanya et al. (2010).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"21 In Chinese, numerals and demonstratives cannot modify nouns directly, and must be accompanied by.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Bikel et al.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":".., Wn ) (obs) P T : Tag assigns (T1 ,.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"AdaBoost was first introduced in (Freund and Schapire 97); (Schapire and Singer 98) gave a generalization of AdaBoost which we will use in this paper.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"RandLM\u2019s stupid backoff variant stores counts instead of probabilities and backoffs.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"am 11.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We have not explored this strategy.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The ATB has a much higher fraction of nuclei per tree, and a higher type-level error rate.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Le m\u00b4edicament de r\u00b4ef\u00b4erence de Silapo est EPREX\/ERYPO, qui contient de l\u2019\u00b4epo\u00b4etine alfa.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Consequently, we cut their evidence values in half to lessen their influence.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For each domain, we created a blind test set by manually annotating 40 doc uments with anaphoric chains, which represent sets of m3 (S) = ) X \u00e2\u0088\u00a9Y =S 1 \u00e2\u0088\u0092 ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) (1) noun phrases that are coreferent (as done for MUC6 (MUC6 Proceedings, 1995)).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Generalizing state minimization, the model could also provide explicit bounds on probability for both backward and forward extension.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We refer to different readings as different analyses whereby the segments are deterministic given the sequence of PoS tags.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Furthermore, the combination of pruning and vertical markovization of the grammar outperforms the Oracle results reported by Cohen and Smith.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"As can be seen in the example, the first two phrases have a different order of NE names from the last two, so we can determine that the last two phrases represent a reversed relation.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"For displaying and querying the annoated text, we make use of the Annis Linguistic Database developed in our group for a large research effort (\u00e2\u0080\u0098Sonderforschungsbereich\u00e2\u0080\u0099) revolving around 9 2.7 Information structure.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"For all languages we do not make use of a tagging dictionary.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Further, the probing hash table does only one random lookup per query, explaining why it is faster on large data.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"For example, CFG's cannot produce trees of the form shown in Figure 1 in which there are nested dependencies between S and NP nodes appearing on the spine of the tree.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The probabilities are incorporated into the DempsterShafer model using Equation 1.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Conditioned on T , features of word types W are drawn.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"As already pointed out in Section 2.4, current theories diverge not only on the number and definition of relations but also on apects of structure, i.e., whether a tree is sufficient as a representational device or general graphs are required (and if so, whether any restrictions can be placed on these graph\u00e2\u0080\u0099s structures \u00e2\u0080\u0094 cf.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Recently, lattices have been used successfully in the parsing of Hebrew (Tsarfaty, 2006; Cohen and Smith, 2007), a Semitic language with similar properties to Arabic.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"May 1995).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Thus, the arc from je to jedna will be labeled 5b\u2193 (to indicate that there is a syntactic head below it).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"There are 13 types of extensions needed to describe the verbgroup reordering.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"As a result, its POS tag needs to be induced in the \u201cNo LP\u201d case, while the 11A word level paired-t-test is significant at p < 0.01 for Danish, Greek, Italian, Portuguese, Spanish and Swedish, and p < 0.05 for Dutch. correct tag is available as a constraint feature in the \u201cWith LP\u201d case.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The hypothesized parse is then the set of constituents that are likely (P > 0.5) to be in the parse according to this model.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Preprocessing the raw trees improves parsing performance considerably.9 We first discard all trees dominated by X, which indicates errors and non-linguistic text.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For example, if FCC and Federal Communications Commission are both found in a document, then Federal has A begin set to 1, Communications has A continue set to 1, Commission has A end set to 1, and FCC has A unique set to 1.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"As observed by Kahane et al. (1998), any (nonprojective) dependency graph can be transformed into a projective one by a lifting operation, which replaces each non-projective arc wj wk by a projective arc wi \u2014* wk such that wi \u2014*\u2217 wj holds in the original graph.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"8 Volume 1, English language, 19961997, Format version 1, correction level 0 An ap ho r T e r r o r i s m R e c Pr F D i s a s t e r s R e c Pr F De f. NP s Pro no uns .43 .79 .55 .50 .72 .59 .42 .91 .58 .42 .82 .56 Tot al .46 .76 .57 .42 .87 .57 Table 2: General Knowledge Sources Table 4: Individual Performance of KSs for Terrorism Table 3: General + Contextual Role Knowledge Sources larger MUC4 and Reuters corpora.9 4.2 Experiments.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The DL-CoTrain algorithm can be motivated as being a greedy method of satisfying the above 2 constraints.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"36.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Other kinds of productive word classes, such as company names, abbreviations (termed fijsuolxie3 in Mandarin), and place names can easily be 20 Note that 7 in E 7 is normally pronounced as leO, but as part of a resultative it is liao3..","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Table 3: Dev set frequencies for the two most significant discourse markers in Arabic are skewed toward analysis as a conjunction.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(Berland and Charniak 99) describe a method for extracting parts of objects from wholes (e.g., &quot;speedometer&quot; from &quot;car&quot;) from a large corpus using hand-crafted patterns.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The contextual role knowledge had the greatest impact on pronouns: +13% recall for terrorism and +15% recall for disasters, with a +1% precision gain in terrorism and a small precision drop of -3% in disasters.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We evaluated the results based on two metrics.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The performance was 80.99% recall and 61.83% precision.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We can 5 Recall that precision is defined to be the number of correct hits divided by the total number of items.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(Blum and Mitchell 98) offer a promising formulation of redundancy, also prove some results about how the use of unlabeled examples can help classification, and suggest an objective function when training with unlabeled examples.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Chris Dyer integrated the code into cdec.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The new algorithm, which we call CoBoost, uses labeled and unlabeled data and builds two classifiers in parallel.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Others depend upon various lexical heuris\u00c2\u00ad tics: for example Chen and Liu (1992) attempt to balance the length of words in a three-word window, favoring segmentations that give approximately equal length for each word.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"In MSA, SVO usually appears in non-matrix clauses.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"To prevent this we &quot;smooth&quot; the confidence by adding a small value, e, to both W+ and W_, giving at = Plugging the value of at from Equ.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Following the setup of Johnson (2007), we use the whole of the Penn Treebank corpus for training and evaluation on English.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"As was explained in the results section, \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d are not desirable keywords in the CC-domain.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The Verbmobil task is an appointment scheduling task.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Combining multiple highly-accurate independent parsers yields promising results.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In the appositive case, the contextual predictor was the head of the modifying appositive (president in the Maury Cooper example); in the second case, the contextual predictor was the preposition together with the noun it modifies (plant_in in the Georgia example).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(a) IDictionary D I D:d\/0.000 B:b\/0.000 B:b\/0.000 ( b ) ( c ) ( d ) I B e s t P a t h ( I d ( I ) o D * ) I cps:nd4.!l(l() Figure 2 An abstract example illustrating the segmentation algorithm.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In this paper, we have presented a new, e\u00c3\u0086cient DP-based search procedure for statistical machine translation.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(7) is at 0 when: 1) Vi : sign(gi (xi)) = sign(g2 (xi)); 2) Ig3(xi)l oo; and 3) sign(gi (xi)) = yi for i = 1, , m. In fact, Zco provides a bound on the sum of the classification error of the labeled examples and the number of disagreements between the two classifiers on the unlabeled examples.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In this paper k = 3 (the three labels are person, organization, location), and we set a = 0.1.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":".., Wn ) (obs) P T : Tag assigns (T1 ,.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The most popular approach to dealing with seg\u00c2\u00ad mentation ambiguities is the maximum matching method, possibly augmented with further heuristics.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Unfortunately, Yarowsky's method is not well understood from a theoretical viewpoint: we would like to formalize the notion of redundancy in unlabeled data, and set up the learning task as optimization of some appropriate objective function.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"In all cases, the key is collapsed to its 64-bit hash.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The property of semilinearity is concerned only with the occurrence of symbols in strings and not their order.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Much could be done to further reduce memory consumption.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The natural baseline (baseline) outperforms the pure IN system only for EMEA\/EP fren.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Nodes are annotated by the name of the composition operation used at that step in the derivation.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"At the phrasal level, we remove all function tags and traces.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Unsupervised Learning of Contextual Role Knowledge for Coreference Resolution","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"There is no global pruning.","target_text":"Translation errors are reported in terms of multireference word error rate (mWER) and subjective sentence error rate (SSER).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In this specific case, as these two titles could fill the same column of an IE table, we regarded them as paraphrases for the evaluation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Clearly the percentage of productively formed words is quite small (for this particular corpus), meaning that dictionary entries are covering most of the 15 GR is .73 or 96%..","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"(Charniak et al., 1996).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The terrorism examples reflect fairly obvious relationships: people who are murdered are killed; agents that \u00e2\u0080\u009creport\u00e2\u0080\u009d things also \u00e2\u0080\u009cadd\u00e2\u0080\u009d and \u00e2\u0080\u009cstate\u00e2\u0080\u009d things; crimes that are \u00e2\u0080\u009cperpetrated\u00e2\u0080\u009d are often later \u00e2\u0080\u009ccondemned\u00e2\u0080\u009d.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The first is an evaluation of the system's ability to mimic humans at the task of segmenting text into word-sized units; the second evaluates the proper-name identification; the third measures the performance on morphological analysis.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"3 61.7 38.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In the appositive case, the contextual predictor was the head of the modifying appositive (president in the Maury Cooper example); in the second case, the contextual predictor was the preposition together with the noun it modifies (plant_in in the Georgia example).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Collisions between two keys in the table can be identified at model building time.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The use of weighted transducers in particular has the attractive property that the model, as it stands, can be straightforwardly interfaced to other modules of a larger speech or natural language system: presumably one does not want to segment Chinese text for its own sake but instead with a larger purpose in mind.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Two sets of examples from Gan are given in (1) and (2) (:::::: Gan's Appendix B, exx.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Foreign names are usually transliterated using hanzi whose sequential pronunciation mimics the source language pronunciation of the name.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"These tag distributions are used to initialize the label distributions over the English vertices in the graph.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"95 B a s e li n e ( S e lf t a g ) 70 a l l B i k e l ( v 1 . 2 ) B a s e l i n e ( P r e t a g ) 7 0 a l l G o l d P O S 70 0.7 70 0.801 278 0.7 52 0.794 278 0.7 71 0.804 295 0.7 52 0.796 295 0.7 75 0.808 309 77.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"However, the overall percentage of non-projective arcs is less than 2% in PDT and less than 1% in DDT.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"MITLM 0.4 (Hsu and Glass, 2008) is mostly designed for accurate model estimation, but can also compute perplexity.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"German English Training: Sentences 58 073 Words 519 523 549 921 Words* 418 979 453 632 Vocabulary Size 7939 4648 Singletons 3454 1699 Test-147: Sentences 147 Words 1 968 2 173 Perplexity { 26:5 Table 4: Multi-reference word error rate (mWER) and subjective sentence error rate (SSER) for three different search procedures.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In addition to the automatic methods, AG, GR, and ST, just discussed, we also added to the plot the values for the current algorithm using only dictionary entries (i.e., no productively derived words or names).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The relevant variables are the set of token-level tags that appear before and after each instance of the ith word type; we denote these context pairs with the set {(tb, ta)} and they are contained in t(\u00e2\u0088\u0092i).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Unigram records store probability, backoff, and an index in the bigram table.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"7 Conclusion and Future Work.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The following two error criteria are used in our experiments: mWER: multi-reference WER: We use the Levenshtein distance between the automatic translation and several reference translations as a measure of the translation errors.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Since MUC6, BBN' s Hidden Markov Model (HMM) based IdentiFinder (Bikel et al., 1997) has achieved remarkably good performance.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Process statistics are already collected by the kernel (and printing them has no meaningful impact on performance).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Source sentence words are aligned with hypothesized target sentence words, where the choice of a new source word, which has not been aligned with a target word yet, is restricted1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"On a set of 11 sentence fragments-the A set-where they reported 100% recall and precision for name identification, we had 73% recall and 80% precision.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Although the ultimate goal of the Verbmobil project is the translation of spoken language, the input used for the translation experiments reported on in this paper is the (more or less) correct orthographic transcription of the spoken sentences.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If they are found in a list, then a feature for that list will be set to 1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This is a rather important source of errors in name identifi\u00c2\u00ad cation, and it is not really possible to objectively evaluate a name recognition system without considering the main lexicon with which it is used.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"(f1; ;mg ; l) 2 (f1; ;mg n fl; l1g ; l0) !","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Syntactic decoders, such as cdec (Dyer et al., 2010), build state from null context then store it in the hypergraph node for later extension.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The local feature groups are: Non-Contextual Feature: This feature is set to 1 for all tokens.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"This work was supported in part by the Swedish Research Council (621-2002-4207).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"German city names are replaced by category markers.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"All of these systems were run on data that was not seen during their development.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The Expectation Maximization (EM) algorithm (Dempster, Laird and Rubin 77) is a common approach for unsupervised training; in this section we describe its application to the named entity problem.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The PROBING model can perform optimistic searches by jumping to any n-gram without needing state and without any additional memory.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"In turn we use two sorts of heuristics, orthogonal to one another, to prune segmentation possibilities based on lexical and grammatical constraints.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"As a result, its POS tag needs to be induced in the \u201cNo LP\u201d case, while the 11A word level paired-t-test is significant at p < 0.01 for Danish, Greek, Italian, Portuguese, Spanish and Swedish, and p < 0.05 for Dutch. correct tag is available as a constraint feature in the \u201cWith LP\u201d case.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"In general, the neighborhoods can be more diverse and we allow a soft label distribution over the vertices.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The corpus has been annotated with six different types of information, which are characterized in the following subsections.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The inventions disclosed herein are the subject of a patent application owned by the University of Utah and licensed on an exclusive basis to Attensity Corporation.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"They considered a semi-supervised POS tagging scenario and showed that one can use a graph over trigram types, and edge weights based on distributional similarity, to improve a supervised conditional random field tagger.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Ends with the feminine affix :: p. 4.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"\u00ce\u00b2 is the shared hyperparameter for the tag assignment prior and word feature multinomials.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Person\u00e2\u0080\u009d domain, 618 keywords are found, and in the \u00e2\u0080\u009cCountry \u00e2\u0080\u0093 Country\u00e2\u0080\u009d domain, 303 keywords are found.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"However, it can be noted that the results for the least informative encoding, Path, are almost comparable, while the third encoding, Head, gives substantially worse results for both data sets.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The ATB gives several different analyses to these words to indicate different types of coordination.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thus, the tree sets generated by HG's are similar to those of CFG's, with each node annotated by the operation (concatenation or wrapping) used to combine the headed strings derived by the daughters of Tree Adjoining Grammars, a tree rewriting formalism, was introduced by Joshi, Levy and Takahashi (1975) and Joshi (1983\/85).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The development of automatic scoring methods is an open field of research.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"evaluated to account for the same fraction of the data.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Due to the dramatic fiscal situation in Brandenburg she now surprisingly withdrew legislation drafted more than a year ago, and suggested to decide on it not before 2003.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"BABAR uses a named entity recognizer to identify proper names that refer to people and companies.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"A related point is that mutual information is helpful in augmenting existing electronic dictionaries, (cf.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"This work was supported in part by the National Science Foundation under grant IRI9704240.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"How should the absence of vowels and syntactic markers influence annotation choices and grammar development?","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Floating point values may be stored in the trie exactly, using 31 bits for non-positive log probability and 32 bits for backoff5.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"This suggests that by generalizing the notion of context-freeness in CFG's, we can define a class of grammatical formalisms that manipulate more complex structures.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"This paper, however, provides a comprehensive overview of the data collection effort and its current state.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"They considered a semi-supervised POS tagging scenario and showed that one can use a graph over trigram types, and edge weights based on distributional similarity, to improve a supervised conditional random field tagger.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"An examination of the subjects' bracketings confirmed that these instructions were satisfactory in yielding plausible word-sized units.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Hence, for each token, one of the four features zone-TXT, zone- HL, zone-DATELINE, or zone-DD is set to 1, and the other 3 are set to 0.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This is motivated by taking \u03b2 po(s|t) to be the parameters of a Dirichlet prior on phrase probabilities, then maximizing posterior estimates p(s|t) given the IN corpus.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The computing time is low, since no reordering is carried out.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(1992).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Consequently, we cut their evidence values in half to lessen their influence.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"To resolve the anaphor, we survey the final belief values assigned to each candidate\u00e2\u0080\u0099s singleton set.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In our coreference resolver, we define \u00ce\u00b8 to be the set of all candidate antecedents for an anaphor.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"5.1 Data Sets.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Such discrepancies can be aligned via an intermediate level of PoS tags.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"It reads a file with a list of German connectives, and when a text is opened for annotation, it highlights all the words that show up in this list; these will be all the potential connectives.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"AdaBoost is given access to a weak learning algorithm, which accepts as input the training examples, along with a distribution over the instances.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"But Rehbein and van Genabith (2007) showed that Evalb should not be used as an indication of real difference\u00e2\u0080\u0094 or similarity\u00e2\u0080\u0094between treebanks.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The natural baseline (baseline) outperforms the pure IN system only for EMEA\/EP fren.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"If no candidate satisfies this condition (which is often the case), then the anaphor is left unresolved.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We also removed the context-type feature type when using the CoBoost approach.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"An examination of the subjects' bracketings confirmed that these instructions were satisfactory in yielding plausible word-sized units.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"For a sentence x and a state sequence z, a first order Markov model defines a distribution: (9) where Val(X) corresponds to the entire vocabulary.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The difference between the featureless model (+PRIOR) and our full model (+FEATS) is 13.6% and 7.7% average error reduction on best and median settings respectively.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Diacritics can also be used to specify grammatical relations such as case and gender.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each caseframe, BABAR collects the semantic classes associated with the head nouns of NPs that were extracted by the caseframe.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The first is an evaluation of the system's ability to mimic humans at the task of segmenting text into word-sized units; the second evaluates the proper-name identification; the third measures the performance on morphological analysis.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"It is important to bear in mind, though, that this is not an inherent limitation of the model.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Similarly Figures 1 and 2 show how the isolated constituent precision varies by sentence length and the size of the span of the hypothesized constituent.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"All commentaries have been annotated with rhetorical structure, using RSTTool4 and the definitions of discourse relations provided by Rhetorical Structure Theory (Mann, Thompson 1988).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Chang of Tsinghua University, Taiwan, R.O.C., for kindly providing us with the name corpora.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Before explaining our method in detail, we present a brief overview in this subsection.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The bias of automatic methods in favor of statistical systems seems to be less pronounced on out-of-domain test data.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Clearly this is not the only way to estimate word-frequencies, however, and one could consider applying other methods: in partic\u00c2\u00ad ular since the problem is similar to the problem of assigning part-of-speech tags to an untagged corpus given a lexicon and some initial estimate of the a priori probabilities for the tags, one might consider a more sophisticated approach such as that described in Kupiec (1992); one could also use methods that depend on a small hand-tagged seed corpus, as suggested by one reviewer.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"64 94.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If is not initCaps, then (not-initCaps, ) is set to 1.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In (Bikel et al., 1997) and (Bikel et al., 1999), performance was plotted against training data size to show how performance improves with training data size.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Pseudo-code describing the generalized boosting algorithm of Schapire and Singer is given in Figure 1.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For example, if Barry is not in commonWords and is found in the list of person first names, then the feature PersonFirstName will be set to 1.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":", Sun day, then the feature DayOfTheWeek is set to 1.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Therefore, performance is more closely tied to the underlying data structure than to the cache.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This essentially means that their model does not result in a true joint inference, as executions for different tasks involve tuning a parameter separately.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The algorithm builds two classifiers in parallel from labeled and unlabeled data.","label":0},{"source_text":"Here we present two algorithms.","target_text":"This intuition is born out by the experimental results.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Unlike Kahane et al. (1998), we do not regard a projectivized representation as the final target of the parsing process.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Such constraints are derived from training data, expressing some relationship between features and outcome.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Although this is not a precise criterion, most cases we evaluated were relatively clear-cut.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The approach uses both spelling and contextual rules.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We define a symmetric similarity function K(uZ7 uj) over two foreign language vertices uZ7 uj E Vf based on the co-occurrence statistics of the nine feature concepts given in Table 1.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The major problem for our seg\u00c2\u00ad menter, as for all segmenters, remains the problem of unknown words (see Fung and Wu [1994]).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We propose maximizing , where is the sequence of named- entity tags assigned to the words in the sentence , and is the information that can be extracted from the whole document containing . Our system is built on a maximum entropy classifier.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"a classifier.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Figure 4 Input lattice (top) and two segmentations (bottom) of the sentence 'How do you say octopus in Japanese?'.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The method halves the error rate in comparison to a method using the labeled examples alone.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"2 60.7 56.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"To facilitate the comparison of our results to those reported by (Cohen and Smith, 2007) we use their data set in which 177 empty and \u201cmalformed\u201d7 were removed.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We now turn our attention to the recognition of string languages generated by these formalisms (LCFRL's).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The equation for sampling a single type-level assignment Ti is given by, 0.2 0 5 10 15 20 25 30 Iteration Figure 2: Graph of the one-to-one accuracy of our full model (+FEATS) under the best hyperparameter setting by iteration (see Section 5).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Once the lexicon has been drawn, the model proceeds similarly to the standard token-level HMM: Emission parameters \u00ce\u00b8 are generated conditioned on tag assignments T . We also draw transition parameters \u00cf\u0086.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The horizontal axis in this plot represents the most significant dimension, which explains 62% of the variation.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Compared with SRILM, IRSTLM adds several features: lower memory consumption, a binary file format with memory mapping, caching to increase speed, and quantization.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The tokens w are generated by token-level tags t from an HMM parameterized by the lexicon structure.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, it is possible to personify any noun, so in children's stories or fables, i\u00c2\u00a5JJ1l.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The table makes clear that enriching our grammar improves the syntactic performance as well as morphological disambiguation (segmentation and POS tagging) accuracy.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Table 1 briefly describes the seven syntactic heuristics used by BABAR to resolve noun phrases.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The precision and recall measures (described in more detail in Section 3) used in evaluating Treebank parsing treat each constituent as a separate entity, a minimal unit of correctness.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"4.1 Local Features.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Our System Wang, Li, and Chang a. 1\\!f!IP Eflltii \/1\\!f!J:P $1til I b. agm: I a m: c. 5 Bf is Bf 1 d. \"*:t: w _t ff 1 \"* :t: w_tff 1 g., , Transliteration\/Translation chen2zhongl-shenl qu3 'music by Chen Zhongshen ' huang2rong2 youlyoul de dao4 'Huang Rong said soberly' zhangl qun2 Zhang Qun xian4zhang3 you2qingl shang4ren2 hou4 'after the county president You Qing had assumed the position' lin2 quan2 'Lin Quan' wang2jian4 'Wang Jian' oulyang2-ke4 'Ouyang Ke' yinl qi2 bu4 ke2neng2 rong2xu3 tai2du2 er2 'because it cannot permit Taiwan Independence so' silfa3-yuan4zhang3 lin2yang2-gang3 'president of the Judicial Yuan, Lin Yanggang' lin2zhangl-hu2 jiangl zuo4 xian4chang3 jie3shuol 'Lin Zhanghu will give an ex\u00c2\u00ad planation live' jin4\/iang3 nian2 nei4 sa3 xia4 de jinlqian2 hui4 ting2zhi3 'in two years the distributed money will stop' gaoltangl da4chi2 ye1zi0 fen3 'chicken stock, a tablespoon of coconut flakes' you2qingl ru4zhu3 xian4fu3 lwu4 'after You Qing headed the county government' Table 5 Performance on morphological analysis.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The difference is due to more careful S-NOM NP NP NP VP VBG :: b NP restoring NP ADJP NN :: b NP NN NP NP ADJP DTJJ ADJP DTJJ NN :: b NP NP NP ADJP ADJP DTJJ J ..i NN :: b NP NP NP ADJP ADJP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ DTJJ J ..i _;\u00ef\u00bf\u00bd PRP J ..i NN _;\u00ef\u00bf\u00bd NP PRP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ J ..i role its constructive effective (b) Stanford (c) Berkeley (d) Bik el (a) Reference Figure 4: The constituent Restoring of its constructive and effective role parsed by the three different models (gold segmentation).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"We also see that the increase in the size of the label sets for Head and Head+Path is far below the theoretical upper bounds given in Table 1.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"5.1 Data Sets.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"It is also worth pointing out a connection with Daum\u00b4e\u2019s (2007) work that splits each feature into domain-specific and general copies.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"One of the strengths of the DempsterShafer model is its natural ability to recognize when several credible hypotheses are still in play.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The form mnh itself can be read as at least three different verbs (\u201ccounted\u201d, \u201cappointed\u201d, \u201cwas appointed\u201d), a noun (\u201ca portion\u201d), and a possessed noun (\u201cher kind\u201d).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"For all other recursive NPs, we add a common annotation to the POS tag of the head (recursiveNPHead).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Since our objective is to compare distributions of bracketing discrepancies, we do not use heuristics to prune the set of nuclei.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We are also grateful to Markus Dickinson, Ali Farghaly, Nizar Habash, Seth Kulick, David McCloskey, Claude Reichard, Ryan Roth, and Reut Tsarfaty for constructive discussions.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This suggests that the backoff model is as reasonable a model as we can use in the absence of further information about the expected cost of a plural form.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(For some recent corpus-based work on Chinese abbreviations, see Huang, Ahrens, and Chen [1993].)","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"It is likely that the more complex cases, where path information could make a difference, are beyond the reach of the parser in most cases.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"First, from a large corpus, we extract all the NE instance pairs.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The choice of the genre commentary resulted from the fact that an investigation of rhetorical structure, its interaction with other aspects of discourse structure, and the prospects for its automatic derivation are the key motivations for building up the corpus.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Evaluation metrics used are Attachment Score (AS), i.e. the proportion of tokens that are attached to the correct head, and Exact Match (EM), i.e. the proportion of sentences for which the dependency graph exactly matches the gold standard.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"36.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"BABAR uses information extraction patterns to identify contextual roles and creates four contextual role knowledge sources using unsupervised learning.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"\u00e2\u0080\u009cAgree\u00e2\u0080\u009d is a subject control verb, which dominates another verb whose subject is the same as that of \u00e2\u0080\u009cagree\u00e2\u0080\u009d; the latter verb is generally the one of interest for extraction.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"92 76.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"By using the output of a hand-coded system such as Proteus, MENE can improve its performance, and can even outperform IdentiFinder (Borthwick, 1999).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"One possibility is to use n-grams based on mutual information.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The results in Table 2 were achieved on the development set.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Every token is independent of the others, and the sentence lattice is in fact a concatenation of smaller lattices, one for each token.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In this paper, we describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","label":1},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"It is unique, agrees with the maximum-likelihood distribution, and has the exponential form (Della Pietra et al., 1997): where refers to the outcome, the history (or context), and is a normalization function.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This is manifest in the lexical choices but 1 www.coli.unisb.de\/\u00e2\u0088\u00bcthorsten\/tnt\/ Dagmar Ziegler is up to her neck in debt.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Evaluation of the Segmentation as a Whole.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"A corpus of German newspaper commentaries has been assembled at Potsdam University, and annotated with different linguistic information, to different degrees.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Chunking is not enough to find such relationships.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Simply assigning to each word its most frequent associated tag in a corpus achieves 94.6% accuracy on the WSJ portion of the Penn Treebank.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Recent research has demonstrated that incorporating this sparsity constraint improves tagging accuracy.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":", Sun day, then the feature DayOfTheWeek is set to 1.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The first m pairs have labels yi, whereas for i = m + 1, , n the pairs are unlabeled.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"97 81.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For example, the independence assumptions mean that the model fails to capture the dependence between specific and more general features (for example the fact that the feature full.-string=New_York is always seen with the features contains (New) and The baseline method tags all entities as the most frequent class type (organization). contains (York) and is never seen with a feature such as contains (Group) ).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Here, we process only full-form words within the translation procedure.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Wu and Fung introduce an evaluation method they call nk-blind.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Figure 1 shows sample sentences from these domains, which are widely divergent.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Prague Dependency Treebank (Haji\u02c7c et al., 2001b), Danish Dependency Treebank (Kromann, 2003), and the METU Treebank of Turkish (Oflazer et al., 2003), which generally allow annotations with nonprojective dependency structures.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"It is formally straightforward to extend the grammar to include these names, though it does increase the likelihood of overgeneration and we are unaware of any working systems that incorporate this type of name.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"JA DE G O L D G R AS S SI C K NE SS DE AT H R A T 14.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In this case, this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This makes memory usage comparable to our PROBING model.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In this paper, we make a simplifying assumption of one-tag-per-word.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Find keywords for each NE pair The keywords are found for each NE category pair.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Since foreign names can be of any length, and since their original pronunciation is effectively unlimited, the identi\u00c2\u00ad fication of such names is tricky.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We evaluated BABAR on two domains: terrorism and natural disasters.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"2 61.7 64.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For example, co-occurring caseframes may reflect synonymy (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> abducted\u00e2\u0080\u009d) or related events (e.g., \u00e2\u0080\u009c<patient> kidnapped\u00e2\u0080\u009d and \u00e2\u0080\u009c<patient> released\u00e2\u0080\u009d).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Somewhat surprisingly, there do not appear to be large systematic differences between linear and MAP combinations.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Since our goal is to perform well under these measures we will similarly treat constituents as the minimal substructures for combination.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"A compatible view is presented by Charniak et al. (1996) who consider the kind of probabilities a generative parser should get from a PoS tagger, and concludes that these should be P(w|t) \u201cand nothing fancier\u201d.3 In our setting, therefore, the Lattice is not used to induce a probability distribution on a linear context, but rather, it is used as a common-denominator of state-indexation of all segmentations possibilities of a surface form.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"La ng ua ge # To ke ns # W or d Ty pe s # Ta gs E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 1 1 7 3 7 6 6 9 4 3 8 6 2 0 3 5 6 8 6 9 9 6 0 5 2 0 6 6 7 8 8 9 3 3 4 1 9 1 4 6 7 4 9 2 0 6 1 8 3 5 6 2 8 3 9 3 7 2 3 2 5 2 8 9 3 1 1 6 4 5 8 2 0 0 5 7 4 5 2 5 1 2 5 4 2 2 4 7 4 1 Table 2: Statistics for various corpora utilized in experiments.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In the case of adverbial reduplication illustrated in (3b) an adjective of the form AB is reduplicated as AABB.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"4.1 Local Features.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"IRSTLM and BerkeleyLM use this state function (and a limit of N \u22121 words), but it is more strict than necessary, so decoders using these packages will miss some recombination opportunities.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We smooth Prf(p \u2014* (s, p)) for rare and OOV segments (s E l, l E L, s unseen) using a \u201cper-tag\u201d probability distribution over rare segments which we estimate using relative frequency estimates for once-occurring segments.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Both authors are members of the Center for Language and Speech Processing at Johns Hopkins University.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"If the parse contains productions from outside our grammar the machine has no direct method for handling them (e.g. the resulting database query may be syntactically malformed).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"A typical situation is shown in Figure 1.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Because the Bikel parser has been parameter- ized for Arabic by the LDC, we do not change the default model settings.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"4 53.7 43.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"M must enter a universal state and check that each of the k constant substrings are in the appropriate place (as determined by the contents of the first 2k work tapes) on the input tape.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Wu and Fung introduce an evaluation method they call nk-blind.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This leads to a linear combination of domain-specific probabilities, with weights in [0, 1], normalized to sum to 1.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Sequence of Initial Caps (SOIC): In the sentence Even News Broadcasting Corp., noted for its accurate reporting, made the erroneous announcement., a NER may mistake Even News Broadcasting Corp. as an organization name.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"However, it is robust, efficient, and easy to implement.4 To perform the maximization in (7), we used the popular L-BFGS algorithm (Liu and Nocedal, 1989), which requires gradient information.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"While the first three models get three to four tags wrong, our best model gets only one word wrong and is the most accurate among the four models for this example.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The question of what soft function to pick, and how to design' algorithms which optimize it, is an open question, but appears to be a promising way of looking at the problem.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The method just described segments dictionary words, but as noted in Section 1, there are several classes of words that should be handled that are not found in a standard dictionary.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"This property reflects an important aspect of the underlying linguistic theory associated with the formalism.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This WFST is then summed with the WFST implementing the dictionary and morphological rules, and the transitive closure of the resulting transducer is computed; see Pereira, Riley, and Sproat (1994) for an explanation of the notion of summing WFSTs.12 Conceptual Improvements over Chang et al.'s Model.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"This design leads to a significant reduction in the computational complexity of training and inference.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"For instance, on Spanish, the absolute gap on median performance is 10%.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"(b) POS tagging accuracy is lowest for maSdar verbal nouns (VBG,VN) and adjectives (e.g., JJ).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We are also grateful to Markus Dickinson, Ali Farghaly, Nizar Habash, Seth Kulick, David McCloskey, Claude Reichard, Ryan Roth, and Reut Tsarfaty for constructive discussions.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Equation 2 is an estimate of the conditional probability of the label given the feature, P(yjx).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"There have been other kinds of efforts to discover paraphrase automatically from corpora.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"01 75.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"E.g. when 'Zahnarzttermin' is aligned to dentist's, the extended lexicon model might learn that 'Zahnarzttermin' actuallyhas to be aligned to both dentist's and ap pointment.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"During coreference resolution, the caseframe network provides evidence that an anaphor and prior noun phrase might be coreferent.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"This approach is similar to BABAR in that they both acquire knowledge from earlier resolutions.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Then, it can be verified that We can now derive the CoBoost algorithm as a means of minimizing Zco.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Table 2 Similarity matrix for segmentation judgments.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"These sequence models-based approaches commonly treat token-level tag assignment as the primary latent variable.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Reference resolution involves finding words that co-refer to the same entity.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Given this similarity function, we define a nearest neighbor graph, where the edge weight for the n most similar vertices is set to the value of the similarity function and to 0 for all other vertices.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Feature selection is implemented using a feature cutoff: features seen less than a small count during training will not be used.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Again, the monotone search performs worst.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"of Tokens MENERGI 318 160,000 200 180,000 IdentiFinder \u00e2\u0080\u0093 650,000 \u00e2\u0080\u0093 790,000 MENE \u00e2\u0080\u0093 \u00e2\u0080\u0093 350 321,000 Table 4: Training Data MUC7 test accuracy.2 For MUC6, the reduction in error due to global features is 27%, and for MUC7,14%.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"This design leads to a significant reduction in the computational complexity of training and inference.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"In this paper we have proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Due to the dramatic fiscal situation in Brandenburg she now surprisingly withdrew legislation drafted more than a year ago, and suggested to decide on it not before 2003.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For example, if X and Y are coreferent, then both X and Y are considered to co-occur with the caseframe that extracts X as well as the caseframe that extracts Y. We will refer to the set of nouns that co-occur with a caseframe as the lexical expectations of the case- frame.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"First we mark any node that dominates (at any level) a verb sider POS tags when pre-terminals are the only intervening nodes between the nucleus and its bracketing (e.g., unaries, base NPs).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The model described here thus demonstrates great potential for use in widespread applications.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In (1) the sequencema3lu4 cannot be resolved locally, but depends instead upon broader context; similarly in (2), the sequence :::tcai2neng2 cannot be resolved locally: 1.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"I = 1X21 N and N is a &quot;medium&quot; sized number so that it is feasible to collect 0(N) unlabeled examples.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"It is not immediately obvious how to formulate an equivalent to equation (1) for an adapted TM, because there is no well-defined objective for learning TMs from parallel corpora.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Zone: MUC data contains SGML tags, and a document is divided into zones (e.g., headlines and text zones).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Smith estimates Lotus will make profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Conversely, the lattice parser requires no linguistic resources and produces segmentations of comparable quality.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Here NO counts the number of hypothesized constituents in the development set that match the binary predicate specified as an argument.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"18 77.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For every sequence of initial capitalized words, its longest substring that occurs in the same document as a sequence of initCaps is identified.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"(f1; ;mg n fl1; l2; l3g ;m) German to English the monotonicity constraint is violated mainly with respect to the German verbgroup.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A greedy algorithm (or maximum-matching algorithm), GR: proceed through the sentence, taking the longest match with a dictionary entry at each point.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"1 is given in Fig.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, we have reason to doubt Chang et al.'s performance claims.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"na me =>1 ha nzi fa mi ly 2 ha nzi gi ve n 3.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The intuition here is that the role of a discourse marker can usually be de 9 Both the corpus split and pre-processing code are avail-.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"2.1 Inverted Alignments.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The trie data structure is commonly used for language modeling.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This group consists of (3 total number of possible zones) features.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The problem with these styles of evaluation is that, as we shall demonstrate, even human judges do not agree perfectly on how to segment a given text.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The present proposal falls into the last group.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In these examples, the names identified by the two systems (if any) are underlined; the sentence with the correct segmentation is boxed.19 The differences in performance between the two systems relate directly to three issues, which can be seen as differences in the tuning of the models, rather than repre\u00c2\u00ad senting differences in the capabilities of the model per se.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We have checked if there are similar verbs in other major domains, but this was the only one.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This is consistent with the nature of these two settings: log-linear combination, which effectively takes the intersection of IN and OUT, does relatively better on NIST, where the domains are broader and closer together.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Historically, Arabic grammar has identified two sentences types: those that begin with a nominal (\u00ef\u00bf\u00bd '.i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"32 81.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"To this end, we construct a bilingual graph over word types to establish a connection between the two languages (\u00a73), and then use graph label propagation to project syntactic information from English to the foreign language (\u00a74).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"5.2 Setup.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Given counts cn1 where e.g. c1 is the vocabulary size, total memory consumption, in bits, is Our PROBING data structure places all n-grams of the same order into a single giant hash table.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"To be short, we omit the target words e; e0 in the formulation of the search hypotheses.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"3.3 Evaluation Results.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"This approach leads to a search procedure with complexity O(E3 J4).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"If the expression is longer or complicated (like \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d and \u00e2\u0080\u009cA\u00e2\u0080\u0099s purchase of B\u00e2\u0080\u009d), it is called \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d, i.e. a set of phrases which express the same thing or event.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"10.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For example, if a token starts with a capital letter and ends with a period (such as Mr.), then the feature InitCapPeriod is set to 1, etc. First Word: This feature group contains only one feature firstword.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"As can be seen in the example, the first two phrases have a different order of NE names from the last two, so we can determine that the last two phrases represent a reversed relation.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"constitute names, since we have only their segmentation, not the actual classification of the segmented words.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The authors provided us with a ratio between TPT and SRI under different conditions. aLossy compression with the same weights. bLossy compression with retuned weights. ditions make the value appropriate for estimating repeated run times, such as in parameter tuning.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The substructures that are unanimously hypothesized by the parsers should be preserved after combination, and the combination technique should not foolishly create substructures for which there is no supporting evidence.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This suggests a direct parallel to (1): where \u02dcp(s, t) is a joint empirical distribution extracted from the IN dev set using the standard procedure.2 An alternative form of linear combination is a maximum a posteriori (MAP) combination (Bacchiani et al., 2004).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"This problem arises because our keywords consist of only one word.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The local features used are similar to those used in BBN' s IdentiFinder (Bikel et al., 1999) or MENE (Borthwick, 1999).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"While there are other obstacles to completing this idea, we believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We performed three experiments to evaluate our techniques.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, we have reason to doubt Chang et al.'s performance claims.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"However, to classify a token , while Borthwick uses tokens from to (from two tokens before to two tokens after ), we used only the tokens , , and . Even with local features alone, MENERGI outperforms MENE (Borthwick, 1999).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Once the lexicon has been drawn, the model proceeds similarly to the standard token-level HMM: Emission parameters \u00ce\u00b8 are generated conditioned on tag assignments T . We also draw transition parameters \u00cf\u0086.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"As noted for the perplexity task, we do not expect cache to grow substantially with model size, so RandLM remains a low-memory option.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The natural language processing community is in the strong position of having many available approaches to solving some of its most fundamental problems.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"When two partial hypotheses have equal state (including that of other features), they can be recombined and thereafter efficiently handled as a single packed hypothesis.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"To date we have not done a separate evaluation of foreign-name recognition.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Tables 4 and 5 show BABAR\u00e2\u0080\u0099s performance when just one contextual role knowledge source is used at a time.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Two of the Mainlanders also cluster close together but, interestingly, not particularly close to the Taiwan speakers; the third Mainlander is much more similar to the Taiwan speakers.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"So we decided to use semantic class information only to rule out candidates.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"If two phrases can be used to express the same relationship within an information extraction application (\u00e2\u0080\u009cscenario\u00e2\u0080\u009d), these two phrases are paraphrases.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The anaphor and antecedent appear in boldface.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"2 62.6 45.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Mai.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Our second contribution is to apply instance weighting at the level of phrase pairs.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"If we remove this sample from the evaluation, then the ATB type-level error rises to only 37.4% while the n-gram error rate increases to 6.24%.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Graph construction for structured prediction problems such as POS tagging is non-trivial: on the one hand, using individual words as the vertices throws away the context necessary for disambiguation; on the other hand, it is unclear how to define (sequence) similarity if the vertices correspond to entire sentences.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We then discuss how we adapt and generalize a boosting algorithm, AdaBoost, to the problem of named entity classification.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"AdaBoost was first introduced in (Freund and Schapire 97); (Schapire and Singer 98) gave a generalization of AdaBoost which we will use in this paper.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The 3rd block contains the mixture baselines.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"(3)In sentence (1), McCann can be a person or an orga nization.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The second modification is more important, and is discussed in the next section.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Je voudrais pr\u00b4eciser, a` l\u2019adresse du commissaire Liikanen, qu\u2019il n\u2019est pas ais\u00b4e de recourir aux tribunaux nationaux.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"This paper proposes a simple and effective tagging method that directly models tag sparsity and other distributional properties of valid POS tag assignments.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This is because our corpus is not annotated, and hence does not distinguish between the various words represented by homographs, such as, which could be \/adv jiangl 'be about to' orInc jiang4 '(military) general'-as in 1j\\xiao3jiang4 'little general.'","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Here NO counts the number of hypothesized constituents in the development set that match the binary predicate specified as an argument.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"collected too.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We used C = 1.0 as the L2 regularization constant in (Eq.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We are also grateful to Markus Dickinson, Ali Farghaly, Nizar Habash, Seth Kulick, David McCloskey, Claude Reichard, Ryan Roth, and Reut Tsarfaty for constructive discussions.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This representation gives ir, an appropriate morphological decomposition, pre\u00c2\u00ad serving information that would be lost by simply listing ir, as an unanalyzed form.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(a) IDictionary D I D:d\/0.000 B:b\/0.000 B:b\/0.000 ( b ) ( c ) ( d ) I B e s t P a t h ( I d ( I ) o D * ) I cps:nd4.!l(l() Figure 2 An abstract example illustrating the segmentation algorithm.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"I \u00e2\u0080\u00a2 JAPANS :rl4 .\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\)\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"o\u00c2\u00b7'\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\:J\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7 \u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7'\\; . '.:: ..........0 6.51 9.51 : jj \/ JAPANESE OCTOPUS 10\u00c2\u00b728i\u00c2\u00a3 :_nc HOW SAY f B :rl4 :il: :wen2 t '- \u00e2\u0080\u00a2 :zhang!","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"A contextual role represents the role that a noun phrase plays in an event or relationship.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Text generation, or at least the two phases of text planning and sentence planning, is a process driven partly by well-motivated choices (e.g., use this lexeme X rather than that more colloquial near-synonym Y ) and partly by con tation like that of PCC can be exploited to look for correlations in particular between syntactic structure, choice of referring expressions, and sentence-internal information structure.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The foreign language vertices (denoted by Vf) correspond to foreign trigram types, exactly as in Subramanya et al. (2010).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The general idea for the knowledge- based part is to have the system use as much information as it can find at its disposal to produce a target representation as specific as possible and as underspecified as necessary.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"But we follow the more direct adaptation of Evalb suggested by Tsarfaty (2006), who viewed exact segmentation as the ultimate goal.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"That is, we can use the discourse parser on PCC texts, emulating for instance a \u00e2\u0080\u009cco-reference oracle\u00e2\u0080\u009d that adds the information from our co-reference annotations.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Figure 4 shows the seven general knowledge sources (KSs) that represent features commonly used for coreference resolution.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"35 76.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In this case, Maury Cooper is extracted.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We evaluate our approach on seven languages: English, Danish, Dutch, German, Portuguese, Spanish, and Swedish.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Given parameter estimates, the label for a test example x is defined as We should note that the model in equation 9 is deficient, in that it assigns greater than zero probability to some feature combinations that are impossible.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"All three models evaluated in this paper incorrectly analyze the constituent as iDafa; none of the models attach the attributive adjectives properly.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The use of ILP in learning the desired grammar significantly increases the computational complexity of this method.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We evaluate our model on seven languages exhibiting substantial syntactic variation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"One annotator suggested that this was the case for as much as 10% of our test sentences.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In this case, this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Here, \u00e2\u0080\u009cEG\u00e2\u0080\u009d represents \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We present two algorithms.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Performance typically stabilizes across languages after only a few number of iterations.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"However, the point of RandLM is to scale to even larger data, compensating for this loss in quality.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Search hypotheses are processed separately according to their coverage set C. The best scored hypothesis for each coverage set is computed: QBeam(C) = max e;e0 ;S;j Qe0 (e; S; C; j) The hypothesis (e0; e; S; C; j) is pruned if: Qe0 (e; S; C; j) < t0 QBeam(C); where t0 is a threshold to control the number of surviving hypotheses.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"This alters generation of T as follows: n P (T |\u00cf\u0088) = n P (Ti|\u00cf\u0088) i=1 Note that this distribution captures the frequency of a tag across word types, as opposed to tokens.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We map the ATB morphological analyses to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags for all experiments.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"When a collision occurs, linear probing places the entry to be inserted in the next (higher index) empty bucket, wrapping around as necessary.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Thus, we feel fairly confident that for the examples we have considered from Gan's study a solution can be incorporated, or at least approximated, within a finite-state framework.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"9 www.ling.unipotsdam.de\/sfb\/ Figure 2: Screenshot of Annis Linguistic Database 3.3 Symbolic and knowledge-based.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Gazdar (1985) argues this is the appropriate analysis of unbounded dependencies in the hypothetical Scandinavian language Norwedish.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"One obvious application is information extraction.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"37 84.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"However, it is robust, efficient, and easy to implement.4 To perform the maximization in (7), we used the popular L-BFGS algorithm (Liu and Nocedal, 1989), which requires gradient information.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Sometimes, however, these beliefs can be contradictory.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"A token that is allCaps will also be initCaps.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Trying to integrate constituent ordering and choice of referring expressions, (Chiarcos 2003) developed a numerical model of salience propagation that captures various factors of author\u00e2\u0080\u0099s intentions and of information structure for ordering sentences as well as smaller constituents, and picking appropriate referring expressions.10 Chiarcos used the PCC annotations of co-reference and information structure to compute his numerical models for salience projection across the generated texts.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The semantic agreement KS eliminates some candidates, but also provides positive evidence in one case: if the candidate and anaphor both have semantic tags human, company, date, or location that were assigned via NER or the manually labeled dictionary entries.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The hyperparameters \u00ce\u00b1 and \u00ce\u00b2 represent the concentration parameters of the token- and type-level components of the model respectively.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"6One of our experimental settings lacks document boundaries, and we used this approximation in both settings for consistency.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Terrorism systems must distinguish between people who perpetrate a crime and people who are victims of a crime.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This class-based model gives reasonable results: for six radical classes, Table 1 gives the estimated cost for an unseen hanzi in the class occurring as the second hanzi in a double GIVEN name.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"This feature ft incorporates information from the smoothed graph and prunes hidden states that are inconsistent with the thresholded vector tx.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The role that each noun phrase plays in the kidnapping event is key to distinguishing these cases.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The remaining arcs are marked OOV.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Similarly Figures 1 and 2 show how the isolated constituent precision varies by sentence length and the size of the span of the hypothesized constituent.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Search hypotheses are processed separately according to their coverage set C. The best scored hypothesis for each coverage set is computed: QBeam(C) = max e;e0 ;S;j Qe0 (e; S; C; j) The hypothesis (e0; e; S; C; j) is pruned if: Qe0 (e; S; C; j) < t0 QBeam(C); where t0 is a threshold to control the number of surviving hypotheses.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The table shows that the lexicon tag frequency predicated by our full model are the closest to the gold standard.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"On each language we investigate the contribution of each component of our model.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Therefore, performance is more closely tied to the underlying data structure than to the cache.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Given a key k, it estimates the position If the estimate is exact (A[pivot] = k), then the algorithm terminates succesfully.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Search hypotheses are processed separately according to their coverage set C. The best scored hypothesis for each coverage set is computed: QBeam(C) = max e;e0 ;S;j Qe0 (e; S; C; j) The hypothesis (e0; e; S; C; j) is pruned if: Qe0 (e; S; C; j) < t0 QBeam(C); where t0 is a threshold to control the number of surviving hypotheses.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Table 2 shows results for both settings and all methods described in sections 2 and 3.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"There has been additional recent work on inducing lexicons or other knowledge sources from large corpora.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"As you can see in the figure, the accuracy for the domain is quite high except for the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set, which contains various expressions representing different relationships for an IE application.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In contrast, our morphological probabilities are based on a unigram, lexeme-based model, and all other (local and non-local) contextual considerations are delegated to the PCFG.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Each parse is converted into a set of constituents represented as a tuples: (label, start, end).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"In the first stage, we run a single step of label propagation, which transfers the label distributions from the English vertices to the connected foreign language vertices (say, Vf\ufffd) at the periphery of the graph.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"If there are too many distinct states, the decoder prunes low-scoring partial hypotheses, possibly leading to a search error.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"IdentiFinder ' 99' s results are considerably better than IdentiFinder ' 97' s. IdentiFinder' s performance in MUC7 is published in (Miller et al., 1998).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"18 77.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"6 Results and Analysis.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Crucially, the conventional orthographic form of MSA text is unvocalized, a property that results in a deficient graphical representation.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"86 78.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(For some recent corpus-based work on Chinese abbreviations, see Huang, Ahrens, and Chen [1993].)","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In considering the recognition of these languages, we were forced to be more specific regarding the relationship between the structures derived by these formalisms and the substrings they span.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"At most one feature in this group will be set to 1.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"For effectively annotating connectives\/scopes, we found that existing annotation tools were not well-suited, for two reasons: \u00e2\u0080\u00a2 Some tools are dedicated to modes of annotation (e.g., tiers), which could only quite un-intuitively be used for connectives and scopes.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Similar behavior is observed when adding features.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"e0; e are the last two target words, C is a coverage set for the already covered source positions and j is the last position visited.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The first 3770 trees of the resulting set then were used for training, and the last 418 are used testing.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"There is no global pruning.","target_text":"A search restriction especially useful for the translation direction from German to English is presented.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We do not adapt the alignment procedure for generating the phrase table from which the TM distributions are derived.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"For rules p : A fpo such that fp is constant function, giving an elementary structure, fp is defined such that fp() = (Si ... xi() where each z is a constant string.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"If (wi, r, wj) E A, we say that wi is the head of wj and wj a dependent of wi.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The cost is computed as follows, where N is the corpus size and f is the frequency: (1) Besides actual words from the base dictionary, the lexicon contains all hanzi in the Big 5 Chinese code\/ with their pronunciation(s), plus entries for other characters that can be found in Chinese text, such as Roman letters, numerals, and special symbols.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Consider first the examples in (2).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The table shows that the lexicon tag frequency predicated by our full model are the closest to the gold standard.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Secondly, some segments in a proposed segment sequence may in fact be seen lexical events, i.e., for some p tag Prf(p \u2014* (s, p)) > 0, while other segments have never been observed as a lexical event before.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"There are clearly eight orthographic words in the example given, but if one were doing syntactic analysis one would probably want to consider I'm to consist of two syntactic words, namely I and am.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For each case- frame, BABAR collects the head nouns of noun phrases that were extracted by the caseframe in the training corpus.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A dynamic programming algorithm is then used to select the sequence of word classes with the highest probability.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Hence, we use the bootstrap resampling method described by Koehn (2004).","label":0},{"source_text":"Here we present two algorithms.","target_text":"Yarowsky-cautious does not separate the spelling and contextual features, but does have a limit on the number of rules added at each stage.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"For example, in TAG's a derived auxiliary tree spans two substrings (to the left and right of the foot node), and the adjunction operation inserts another substring (spanned by the subtree under the node where adjunction takes place) between them (see Figure 3).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Obviously \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is part of the following clause rather than being the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d and the extracted instance makes no sense.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The major problem for all segmentation systems remains the coverage afforded by the dictionary and the lexical rules used to augment the dictionary to deal with unseen words.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The original tag set for the CoNLL-X Dutch data set consists of compounded tags that are used to tag multi-word units (MWUs) resulting in a tag set of over 300 tags.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We use double-circles to indicate the space-delimited token boundaries.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"A number of PCC commentaries will be read by professional news speakers and prosodic features be annotated, so that the various annotation layers can be set into correspondence with intonation patterns.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"18 We are grateful to ChaoHuang Chang for providing us with this set.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The cost of storing these averages, in bits, is Because there are comparatively few unigrams, we elected to store them byte-aligned and unquantized, making every query faster.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We present several variations for the lexical component P (T , W |\u00cf\u0088), each adding more complex pa rameterizations.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Section 5 covers relevant previous work on SMT adaptation, and section 6 concludes.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"However, TRIE partitions storage by n-gram length, so walking the trie reads N disjoint pages.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Cur In order to ground such approaches in linguistic observation and description, a multi-level anno 10 For an exposition of the idea as applied to the task of text planning, see (Chiarcos, Stede 2004).","label":0},{"source_text":"Here we present two algorithms.","target_text":"At each iteration the algorithm increases the number of rules, while maintaining a high level of agreement between the spelling and contextual decision lists.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Past work however, has typically associ n = n P (Ti)P (Wi|Ti) = i=1 1 n K n ated these features with token occurrences, typically in an HMM.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The first step in the learning process is to generate training examples consisting of anaphor\/antecedent resolutions.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Thus, if one wants to segment words-for any purpose-from Chinese sentences, one faces a more difficult task than one does in English since one cannot use spacing as a guide.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"4.1 Corpora.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"and \u00e2\u0080\u009cH\u00e2\u0080\u009d represents \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"an event.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The same numbers were used for each data structure.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The idea of distinguishing between general and domain-specific examples is due to Daum\u00b4e and Marcu (2006), who used a maximum-entropy model with latent variables to capture the degree of specificity.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Though we are not able to calculate their memory usage on our model, results reported in their paper suggest lower memory consumption than TRIE on large-scale models, at the expense of CPU time.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"9 www.ling.unipotsdam.de\/sfb\/ Figure 2: Screenshot of Annis Linguistic Database 3.3 Symbolic and knowledge-based.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"When OUT is large and distinct, its contribution can be controlled by training separate IN and OUT models, and weighting their combination.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Entries landing in the same bucket are said to collide.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Initially, the DempsterShafer model assumes that all hypotheses are equally likely, so it creates a set called \u00ce\u00b8 that includes all hypotheses.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"These alignment models are similar to the concept of hidden Markov models (HMM) in speech recognition.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"La ng ua ge 1T W + P RI O R + F E A T S E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 2 1.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Graph construction for structured prediction problems such as POS tagging is non-trivial: on the one hand, using individual words as the vertices throws away the context necessary for disambiguation; on the other hand, it is unclear how to define (sequence) similarity if the vertices correspond to entire sentences.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In this section, we will explain the algorithm step by step with examples.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Thus, the effects of spontaneous speech are present in the corpus, e.g. the syntactic structure of the sentence is rather less restricted, however the effect of speech recognition errors is not covered.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The gender, number, and scoping KSs eliminate candidates from consideration.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In this subsection, we will report the results of the experiment, in terms of the number of words, phrases or clusters.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We are also grateful to Markus Dickinson, Ali Farghaly, Nizar Habash, Seth Kulick, David McCloskey, Claude Reichard, Ryan Roth, and Reut Tsarfaty for constructive discussions.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For speed, we plan to implement the direct-mapped cache from BerkeleyLM.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"A different but supplementary perspective on discourse-based information structure is taken 11ventionalized patterns (e.g., order of informa by one of our partner projects, which is inter tion in news reports).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"By this assumption, each element x E X can also be represented as (xi, x2) E X1 x X2.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Models that employ this strategy are denoted hsp.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The task is to learn a function from an input string (proper name) to its type, which we will assume to be one of the categories Person, Organization, or Location.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"However, 1 http:\/\/maxent.sourceforge.net 3.2 Testing.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"67 95.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In Section 6 we dis\u00c2\u00ad cuss other issues relating to how higher-order language models could be incorporated into the model.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Moving beyond directly related work, major themes in SMT adaptation include the IR (Hildebrand et al., 2005; L\u00a8u et al., 2007; Zhao et al., 2004) and mixture (Finch and Sumita, 2008; Foster and Kuhn, 2007; Koehn and Schroeder, 2007; L\u00a8u et al., 2007) approaches for LMs and TMs described above, as well as methods for exploiting monolingual in-domain text, typically by translating it automatically and then performing self training (Bertoldi and Federico, 2009; Ueffing et al., 2007; Schwenk and Senellart, 2009).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Yarowsky 95) describes an algorithm for word-sense disambiguation that exploits redundancy in contextual features, and gives impressive performance.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"(f1; ;mg ; l) 2 (f1; ;mg n fl; l1g ; l0) !","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"For the English RST-annotated corpus that is made available via LDC, his corresponding result is 62%.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"4.2 Global Features.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"of Articles No.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"If an unlabeled vertex does not have a path to any labeled vertex, this term ensures that the converged marginal for this vertex will be uniform over all tags, allowing the middle word of such an unlabeled vertex to take on any of the possible tags.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Thus we are interested not in extraction, but actual generation from representations that may be developed to different degrees of granularity.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"For all other recursive NPs, we add a common annotation to the POS tag of the head (recursiveNPHead).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The results described in this paper suggest a characterization of mild context-sensitivity in terms of generalized context-freeness.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"However there is no global pruning.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Mai.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"A possible probabilistic model for assigning probabilities to complex analyses of a surface form may be and indeed recent sequential disambiguation models for Hebrew (Adler and Elhadad, 2006) and Arabic (Smith et al., 2005) present similar models.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Blum and Mitchell 98) offer a promising formulation of redundancy, also prove some results about how the use of unlabeled examples can help classification, and suggest an objective function when training with unlabeled examples.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"2.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models (e.g., (Grosz et al., 1995; Grosz and Sidner, 1998)), syntactic algorithms (e.g., (Hobbs, 1978; Lappin and Le- ass, 1994)), and supervised machine learning systems (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":1},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Each feature group can be made up of many binary features.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"By this assumption, each element x E X can also be represented as (xi, x2) E X1 x X2.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"If the expression is longer or complicated (like \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d and \u00e2\u0080\u009cA\u00e2\u0080\u0099s purchase of B\u00e2\u0080\u009d), it is called \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d, i.e. a set of phrases which express the same thing or event.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The final strong hypothesis, denoted 1(x), is then the sign of a weighted sum of the weak hypotheses, 1(x) = sign (Vii atht(x)), where the weights at are determined during the run of the algorithm, as we describe below.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We can now compare this algorithm to that of (Yarowsky 95).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"There are 13 types of extensions needed to describe the verbgroup reordering.","label":0},{"source_text":"There is no global pruning.","target_text":"The complexity of the quasimonotone search is O(E3 J (R2+LR)).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"It is a relatively frequent word in the domain, but it can be used in different extraction scenarios.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Recently, statistical NERs have achieved results that are comparable to hand-coded systems.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The P (W |T , \u00cf\u0088) term in the lexicon component now decomposes as: n P (W |T , \u00cf\u0088) = n P (Wi|Ti, \u00cf\u0088) i=1 n \u00ef\u00a3\u00ab \u00ef\u00a3\u00b6 tions are not modeled by the standard HMM, which = n \u00ef\u00a3\u00ad n P (v|\u00cf\u0088Ti f )\u00ef\u00a3\u00b8 instead can model token-level frequency.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"This annotation choice weakens splitIN.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, until such standards are universally adopted in evaluating Chinese segmenters, claims about performance in terms of simple measures like percent correct should be taken with a grain of salt; see, again, Wu and Fung (1994) for further arguments supporting this conclusion.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"\u00e2\u0080\u009d, because Bush is the first word, the initial caps might be due to its position (as in \u00e2\u0080\u009cThey put a freeze on . . .","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"1 74.5 56.","label":0},{"source_text":"There is no global pruning.","target_text":"Table 5: Effect of the beam threshold on the number of search errors (147 sentences).","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In practice, we can therefore expect a trade-off such that increasing the amount of information encoded in arc labels will cause an increase in the accuracy of the inverse transformation but a decrease in the accuracy with which the parser can construct the labeled representations.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"To initialize the graph for label propagation we use a supervised English tagger to label the English side of the bitext.7 We then simply count the individual labels of the English tokens and normalize the counts to produce tag distributions over English word types.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Additionally, it works about 3 times as fast as the IBM style search.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This is to allow for fair comparison between the statistical method and GR, which is also purely dictionary-based.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We currently simulate this by crafting a WCFG and feeding it to BitPar.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The first is an evaluation of the system's ability to mimic humans at the task of segmenting text into word-sized units; the second evaluates the proper-name identification; the third measures the performance on morphological analysis.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In our experiment, we set the threshold of the TF\/ITF score empirically using a small development corpus; a finer adjustment of the threshold could reduce the number of such keywords.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"On the other hand, given that all schemes have similar parsing accuracy overall, this means that the Path scheme is the least likely to introduce errors on projective arcs.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Following the convention presented in earlier sections, we assume that each example is an instance pair of the from (xi ,i, x2,) where xj,, E 2x3 , j E 2}.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"This work has been supported as part of the Verbmobil project (contract number 01 IV 601 A) by the German Federal Ministry of Education, Science, Research and Technology and as part of the Eutrans project (ESPRIT project number 30268) by the European Community.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"\u00c2\u00a3 : _ADV: 5.88 If:!","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The increase is generally higher for PDT than for DDT, which indicates a greater diversity in non-projective constructions.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"This class of formalisms have the properties that their derivation trees are local sets, and manipulate objects, using a finite number of composition operations that use a finite number of symbols.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"paper, and is missing 6 examples from the A set.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: The regime gives itself the right...","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Note that hanzi that are not grouped into dictionary words (and are not identified as single\u00c2\u00ad hanzi words), or into one of the other categories of words discussed in this paper, are left unattached and tagged as unknown words.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Specifically, the limit n starts at 5 and increases by 5 at each iteration.)","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Our smoothing procedure takes into account all the aforementioned aspects and works as follows.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"There are two weaknesses in Chang et al.'s model, which we improve upon.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"So the success of the algorithm may well be due to its success in maximizing the number of unlabeled examples on which the two decision lists agree.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"If evidence indicates that hypotheses C and D are less likely than hypotheses A and B, then probabilities are redistributed to reflect the fact that {A, B} is more likely to contain the answer than {C, D}.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Past work however, has typically associ n = n P (Ti)P (Wi|Ti) = i=1 1 n K n ated these features with token occurrences, typically in an HMM.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Also, the argument has been made that machine translation performance should be evaluated via task-based evaluation metrics, i.e. how much it assists performing a useful task, such as supporting human translators or aiding the analysis of texts.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Finally, a DempsterShafer probabilistic model evaluates the evidence provided by the knowledge sources for all candidate antecedents and makes the final resolution decision.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Fall.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The ATB segmentation scheme is one of many alternatives.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"While RST (Mann, Thompson 1988) proposed that a single relation hold between adjacent text segments, SDRT (Asher, Lascarides 2003) maintains that multiple relations may hold simultaneously.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Precision.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We combined evidence from four contextual role knowledge sources with evidence from seven general knowledge sources using a DempsterShafer probabilistic model.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Given around 90,000 unlabeled examples, the methods described in this paper classify names with over 91% accuracy.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The ATB segmentation scheme is one of many alternatives.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"It was filtered to retain the top 30 translations for each source phrase using the TM part of the current log-linear model.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A Person-Prefix-List is compiled in an analogous way.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"KenLM: Faster and Smaller Language Model Queries","label":1},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"It is interesting to note, however, that the ability to produce a bounded number of dependent paths (where two dependent paths can share an unbounded amount of information) does not require machinery as powerful as that used in LFG, FUG and IG's.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The baseline simply retains the original labels for all arcs, regardless of whether they have been lifted or not, and the number of distinct labels is therefore simply the number n of distinct dependency types.2 In the first encoding scheme, called Head, we use a new label d\u2191h for each lifted arc, where d is the dependency relation between the syntactic head and the dependent in the non-projective representation, and h is the dependency relation that the syntactic head has to its own head in the underlying structure.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Denote by g3(x) = Et crithl(x) , j E {1,2} the unthresholded strong-hypothesis (i.e., f3 (x) = sign(gi (x))).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Standard SMT systems have a hierarchical parameter structure: top-level log-linear weights are used to combine a small set of complex features, interpreted as log probabilities, many of which have their own internal parameters and objectives.","label":0},{"source_text":"There is no global pruning.","target_text":"The monotone search performs worst in terms of both error rates mWER and SSER.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The features we used can be divided into 2 classes: local and global.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"A receives a votes, and B receives b votes.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The general-language features have a slight advantage over the similarity features, and both are better than the SVM feature.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"1 55.8 38.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Initially, the DempsterShafer model assumes that all hypotheses are equally likely, so it creates a set called \u00ce\u00b8 that includes all hypotheses.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Interpolation search is therefore a form of binary search with better estimates informed by the uniform key distribution.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We have provided methods for handling certain classes of unknown words, and models for other classes could be provided, as we have noted.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"This causes a problem for reverse trie implementations, including SRILM itself, because it leaves n+1-grams without an n-gram node pointing to them.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Table 4 shows translation results for the three approaches.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In many cases these failures in recall would be fixed by having better estimates of the actual prob\u00c2\u00ad abilities of single-hanzi words, since our estimates are often inflated.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"(S; C; j); Not only the coverage set C and the positions j; j0, but also the verbgroup states S; S0 are taken into account.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Finally, since non-projective constructions often involve long-distance dependencies, the problem is closely related to the recovery of empty categories and non-local dependencies in constituency-based parsing (Johnson, 2002; Dienes and Dubey, 2003; Jijkoun and de Rijke, 2004; Cahill et al., 2004; Levy and Manning, 2004; Campbell, 2004).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Table 1 shows four words \u00e2\u0080\u009c 0 Indeed NN Indeed Saddamwhose unvocalized surface forms 0 an are indistinguishable.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"We respond to this on the one hand with a format for its underspecification (see 2.4) and on the other hand with an additional level of annotation that attends only to connectives and their scopes (see 2.5), which is intended as an intermediate step on the long road towards a systematic and objective treatment of rhetorical structure.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"They cluster NE instance pairs based on the words in the contexts using a bag- of-words method.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Using a treebank grammar, a data-driven lexicon, and a linguistically motivated unknown-tokens handling technique our model outperforms previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The best analysis of the corpus is taken to be the true analysis, the frequencies are re-estimated, and the algorithm is repeated until it converges.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"This vector tx is constructed for every word in the foreign vocabulary and will be used to provide features for the unsupervised foreign language POS tagger.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Each connected path (l1 ... lk) E L corresponds to one morphological segmentation possibility of W. The Parser Given a sequence of input tokens W = w1 ... wn and a morphological analyzer, we look for the most probable parse tree \u03c0 s.t.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Realizing gains in practice can be challenging, however, particularly when the target domain is distant from the background data.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Figure 4 Input lattice (top) and two segmentations (bottom) of the sentence 'How do you say octopus in Japanese?'.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Our work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"A similar maximumlikelihood approach was used by Foster and Kuhn (2007), but for language models only.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"(1998) did make use of information from the whole document.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Note that in our formalism a weakhypothesis can abstain.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"For instance, by altering the emission distribution parameters, Johnson (2007) encourages the model to put most of the probability mass on few tags.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"However, lexically similar NPs usually refer to the same entity in two cases: proper names and existential noun phrases.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"ments contained 322 anaphoric links.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The word joining is done on the basis of a likelihood criterion.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Our coreference resolver also incorporates an existential noun phrase recognizer and a DempsterShafer probabilistic model to make resolution decisions.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The joint morphological and syntactic hypothesis was first discussed in (Tsarfaty, 2006; Tsarfaty and Sima\u2019an, 2004) and empirically explored in (Tsarfaty, 2006).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Historically, Arabic grammar has identified two sentences types: those that begin with a nominal (\u00ef\u00bf\u00bd '.i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The accuracies for link were 73% and 86% on two evaluated domains.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"4 69.0 51.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"On the other hand, we are interested in the application of rhetorical analysis or \u00e2\u0080\u0098discourse parsing\u00e2\u0080\u0099 (3.2 and 3.3), in text generation (3.4), and in exploiting the corpus for the development of improved models of discourse structure (3.5).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"A defining characteristic of MSA is the prevalence of discourse markers to connect and subordinate words and phrases (Ryding, 2005).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In Section 4, we present the performance measures used and give translation results on the Verbmobil task.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"A Brief Introduction to the Chinese Writing System Most readers will undoubtedly be at least somewhat familiar with the nature of the Chinese writing system, but there are enough common misunderstandings that it is as well to spend a few paragraphs on properties of the Chinese script that will be relevant to topics discussed in this paper.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Since every CFL is known to be semilinear (Parikh, 1966), in order to show semilinearity of some language, we need only show the existence of a letter equivalent CFL Our definition of LCFRS's insists that the composition operations are linear and nonerasing.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Furthermore we expect the label distributions on the foreign to be fairly noisy, because the graph constraints have not been taken into account yet.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"This PCFG is incorporated into the Stanford Parser, a factored model that chooses a 1-best parse from the product of constituency and dependency parses.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"For example, syntactic decoders (Koehn et al., 2007; Dyer et al., 2010; Li et al., 2009) perform dynamic programming parametrized by both backward- and forward-looking state.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"To establish a soft correspondence between the two languages, we use a second similarity function, which leverages standard unsupervised word alignment statistics (\u00a73.3).3 Since we have no labeled foreign data, our goal is to project syntactic information from the English side to the foreign side.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"A less canonical representation of segmental morphology is triggered by a morpho-phonological process of omitting the definite article h when occurring after the particles b or l. This process triggers ambiguity as for the definiteness status of Nouns following these particles.We refer to such cases in which the concatenation of elements does not strictly correspond to the original surface form as super-segmental morphology.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Mixing, smoothing, and instance-feature weights are learned at the same time using an efficient maximum-likelihood procedure that relies on only a small in-domain development corpus.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Specifically, the lexicon is generated as: P (T , W |\u00cf\u0088) =P (T )P (W |T ) Word Type Features (FEATS): Past unsupervised POS work have derived benefits from features on word types, such as suffix and capitalization features (Hasan and Ng, 2009; Berg-Kirkpatrick et al.,2010).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"So, it is too costly to make IE technology \u00e2\u0080\u009copen- domain\u00e2\u0080\u009d or \u00e2\u0080\u009con-demand\u00e2\u0080\u009d like IR or QA.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"It may seem surprising to some readers that the interhuman agreement scores reported here are so low.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"After adding a ROOT node to all trees, we train a grammar using six split-and- merge cycles and no Markovization.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Morphological segmentation decisions in our model are delegated to a lexeme-based PCFG and we show that using a simple treebank grammar, a data-driven lexicon, and a linguistically motivated unknown-tokens handling our model outperforms (Tsarfaty, 2006) and (Cohen and Smith, 2007) on the joint task and achieves state-of-the-art results on a par with current respective standalone models.2","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Compared with SRILM, IRSTLM adds several features: lower memory consumption, a binary file format with memory mapping, caching to increase speed, and quantization.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Otherwise, label the training data with the combined spelling\/contextual decision list, then induce a final decision list from the labeled examples where all rules (regardless of strength) are added to the decision list.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"MUC7 has also seen hybrids of statistical NERs and hand-coded systems (Mikheev et al., 1998; Borthwick, 1999), notably Mikheev' s system, which achieved the best performance of 93.39% on the official NE test data.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"A Hebrew surface token may have several readings, each of which corresponding to a sequence of segments and their corresponding PoS tags.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"On the other hand, only very restricted reorderings are necessary, e.g. for the translation direction from Table 2: Coverage set hypothesis extensions for the IBM reordering.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We then extract a set of possible tags tx(y) by eliminating labels whose probability is below a threshold value \u03c4: We describe how we choose \u03c4 in \u00a76.4.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"(2009) also report results on English, but on the reduced 17 tag set, which is not comparable to ours).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"While there might be some controversy about the exact definition of such a tagset, these 12 categories cover the most frequent part-of-speech and exist in one form or another in all of the languages that we studied.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The only way to handle such phenomena within the framework described here is simply to expand out the reduplicated forms beforehand, and incorporate the expanded forms into the lexical transducer.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"\u00e2\u0080\u009d).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"At this step, we will try to link those sets, and put them into a single cluster.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In addition we report for each model its performance on goldsegmented input (GS) to indicate the upper bound 11Overt definiteness errors may be seen as a wrong feature rather than as wrong constituent and it is by now an accepted standard to report accuracy with and without such errors. for the grammars\u2019 performance on the parsing task.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"By this assumption, each element x E X can also be represented as (xi, x2) E X1 x X2.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Minimal perfect hashing is used to find the index at which a quantized probability and possibly backoff are stored.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Somewhat surprisingly, there do not appear to be large systematic differences between linear and MAP combinations.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The cost of storing these averages, in bits, is Because there are comparatively few unigrams, we elected to store them byte-aligned and unquantized, making every query faster.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"As described in Sproat (1995), the Chinese segmenter presented here fits directly into the context of a broader finite-state model of text analysis for speech synthesis.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"4.4 Chinese Personal Names.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We asked six native speakers-three from Taiwan (TlT3), and three from the Mainland (M1M3)-to segment the corpus.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Table 2 shows BABAR\u00e2\u0080\u0099s performance.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"RandLM and SRILM also remove context that will not extend, but SRILM performs a second lookup in its trie whereas our approach has minimal additional cost.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"As is standard, we use a fixed constant K for the number of tagging states.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We often encounter sentences that are highly ambiguous in themselves, without some prior knowledge of the entities concerned.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"In this work, we take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Interpolation search formalizes the notion that one opens a dictionary near the end to find the word \u201czebra.\u201d Initially, the algorithm knows the array begins at b +\u2014 0 and ends at e +\u2014 |A|\u22121.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If is a number string (such as one, two, etc), then the feature NumberString is set to 1.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"02 99.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"As you can see in the figure, the accuracy for the domain is quite high except for the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set, which contains various expressions representing different relationships for an IE application.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"This is not an oversight.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For example, given a sequence F1G1G2, where F1 is a legal single-hanzi family name, and Plural Nouns X g 0 g \"' X X 0 T!i c\"'.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Subsets of partial hypotheses with coverage sets C of increasing cardinality c are processed.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The implementation is In a similar effort, (G\u00c2\u00a8otze 2003) developed a proposal for the theory-neutral annotation of information structure (IS) \u00e2\u0080\u0094 a notoriously difficult area with plenty of conflicting and overlapping terminological conceptions.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"However, while Dienes and Dubey recognize empty categories in a pre-processing step and only let the parser find their antecedents, we use the parser both to detect dislocated dependents and to predict either the type or the location of their syntactic head (or both) and use post-processing only to transform the graph in accordance with the parser\u2019s analysis.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"For a corpus C, let M be the set of tuples \u00e2\u0088\u0097n, l), where n is an n-gram with bracketing label l. If any n appears 6 Generative parsing performance is known to deteriorate with sentence length.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If is seen infrequently during training (less than a small count), then will not be selected as a feature and all features in this group are set to 0.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Note, however, that there might be situations in which Zco in fact increases.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Context from the whole document can be important in classifying a named entity.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Still, for both human and automatic rhetorical analysis, connectives are the most important source of surface information.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"3.1 Maximum Entropy.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The paper is organized as follows: Section 2 explains the different layers of annotation that have been produced or are being produced.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Here, all tokeninternal collocations of tags unseen in our training data are pruned away.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"These operations, as we see below, are restricted to be size preserving (as in the case of concatenation in CFG) which implies that they will be linear and non-erasing.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Nodes are annotated by the name of the composition operation used at that step in the derivation.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We carried out translation experiments in two different settings.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"(2003), which gives 96.8% accuracy on the test set.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The complexity of the algorithm is O(E3 J2 2J), where E is the size of the target language vocabulary.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For the error counts, a range from 0:0 to 1:0 is used.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"German city names are replaced by category markers.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"5.2 Discussion.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses unsupervised learning to acquire this knowledge from plain text without the need for annotated training data.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In the third and final scheme, denoted Path, we keep the extra infor2Note that this is a baseline for the parsing experiment only (Experiment 2).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"This paper is based on work supported in part by DARPA through IBM.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Also, we don\u00e2\u0080\u0099t know how many such paraphrase sets are necessary to cover even some everyday things or events.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"While the paper mentioned a sorted variant, code was never released.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"In this paper we study the problem of using a parallel corpus from a background domain (OUT) to improve performance on a target domain (IN) for which a smaller amount of parallel training material\u2014though adequate for reasonable performance\u2014is also available.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Except our own and MENE + reference resolution, the results in Table 6 are all official MUC7 results.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Arabic sentences of up to length 63 would need to be.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The first term in the objective function is the graph smoothness regularizer which encourages the distributions of similar vertices (large wij) to be similar.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Parent Head Modif er Dir # gold F1 Label # gold F1 NP NP TAG R 946 0.54 ADJP 1216 59.45 S S S R 708 0.57 SBAR 2918 69.81 NP NP ADJ P R 803 0.64 FRAG 254 72.87 NP NP N P R 2907 0.66 VP 5507 78.83 NP NP SBA R R 1035 0.67 S 6579 78.91 NP NP P P R 2713 0.67 PP 7516 80.93 VP TAG P P R 3230 0.80 NP 34025 84.95 NP NP TAG L 805 0.85 ADVP 1093 90.64 VP TAG SBA R R 772 0.86 WHN P 787 96.00 S VP N P L 961 0.87 (a) Major phrasal categories (b) Major POS categories (c) Ten lowest scoring (Collins, 2003)-style dependencies occurring more than 700 times Table 8: Per category performance of the Berkeley parser on sentence lengths \u00e2\u0089\u00a4 70 (dev set, gold segmentation).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In the cotraining case, (Blum and Mitchell 98) argue that the task should be to induce functions Ii and f2 such that So Ii and 12 must (1) correctly classify the labeled examples, and (2) must agree with each other on the unlabeled examples.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Mikheev et al.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Tag set As is standard, for all experiments, we set the number of latent model tag states to the size of the annotated tag set.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"We will briefly discuss this point in Section 3.1.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Both of these analyses are shown in Figure 4; fortunately, the correct analysis is also the one with the lowest cost, so it is this analysis that is chosen.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Sentence pairs are the natural instances for SMT, but sentences often contain a mix of domain-specific and general language.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, management succession systems must distinguish between a person who is fired and a person who is hired.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"To control for the effect of the HSPELL-based pruning, we also experimented with a morphological analyzer that does not perform this pruning.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Though we are not able to calculate their memory usage on our model, results reported in their paper suggest lower memory consumption than TRIE on large-scale models, at the expense of CPU time.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Twentieth-century linguistic work on Chinese (Chao 1968; Li and Thompson 1981; Tang 1988,1989, inter alia) has revealed the incorrectness of this traditional view.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"1).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"When this metric is less than 0.5, we expect to incur more errors' than we will remove by adding those constituents to the parse.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"83 77.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(Blum and Mitchell 98) give an example that illustrates just how powerful the second constraint can be.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We have provided methods for handling certain classes of unknown words, and models for other classes could be provided, as we have noted.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"For developers of Statistical Machine Translation (SMT) systems, an additional complication is the heterogeneous nature of SMT components (word-alignment model, language model, translation model, etc.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"To date we have not done a separate evaluation of foreign-name recognition.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"43 80.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"If there are too many distinct states, the decoder prunes low-scoring partial hypotheses, possibly leading to a search error.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, as we have noted, nothing inherent in the approach precludes incorporating higher-order constraints, provided they can be effectively modeled within a finite-state framework.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We adopted this state-of-the-art model because it makes it easy to experiment with various ways of incorporating our novel constraint feature into the log-linear emission model.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The structure uses linear probing hash tables and is designed for speed.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The Input The set of analyses for a token is thus represented as a lattice in which every arc corresponds to a specific lexeme l, as shown in Figure 1.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"It is natural that the larger the data in the domain, the more keywords are found.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The corpus has been annotated with six different types of information, which are characterized in the following subsections.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We proposed an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In this paper we have argued that Chinese word segmentation can be modeled ef\u00c2\u00ad fectively using weighted finite-state transducers.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Across all languages, +PRIOR consistently outperforms 1TW, reducing error on average by 9.1% and 5.9% on best and median settings respectively.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"First, we learn weights on individual phrase pairs rather than sentences.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"In the cases where isolated constituent precision is larger than 0.5 the affected portion of the hypotheses is negligible.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"While Gan's system incorporates fairly sophisticated models of various linguistic information, it has the drawback that it has only been tested with a very small lexicon (a few hundred words) and on a very small test set (thirty sentences); there is therefore serious concern as to whether the methods that he discusses are scalable.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In this domain the major scenarios involve the things they agreed on, rather than the mere fact that they agreed.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"They contain about 200M words (25M, 110M, 40M and 19M words, respectively).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The accuracies for link were 73% and 86% on two evaluated domains.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"However, there is again local grammatical information that should favor the split in the case of (1a): both .ma3 'horse' and .ma3 lu4 are nouns, but only .ma3 is consistent with the classifier pil, the classifier for horses.21 By a similar argument, the preference for not splitting , lm could be strengthened in (lb) by the observation that the classifier 1'1* tiao2 is consistent with long or winding objects like , lm ma3lu4 'road' but not with,ma3 'horse.'","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, the characterization given in the main body of the text is correct sufficiently often to be useful.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The estimation of the probabilities in the model is carried out as shown in Equation 4.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Section 2 describes our baseline techniques for SMT adaptation, and section 3 describes the instance-weighting approach.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"To our knowledge, ours is the first analysis of this kind for Arabic parsing.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"This concerns on the one hand the basic question of retrieval, i.e. searching for information across the annotation layers (see 3.1).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Each model was able to produce hypotheses for all input sentences.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The substructures that are unanimously hypothesized by the parsers should be preserved after combination, and the combination technique should not foolishly create substructures for which there is no supporting evidence.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The translation search is carried out with the category markers and the city names are resubstituted into the target sentence as a postprocessing step.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"See Section 5.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The dependency graph in Figure 1 satisfies all the defining conditions above, but it fails to satisfy the condition ofprojectivity (Kahane et al., 1998): The arc connecting the head jedna (one) to the dependent Z (out-of) spans the token je (is), which is not dominated by jedna.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Ends with the feminine affix :: p. 4.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In this domain the major scenarios involve the things they agreed on, rather than the mere fact that they agreed.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Borth- wick (1999) successfully made use of other hand- coded systems as input for his MENE system, and achieved excellent results.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The choice of the genre commentary resulted from the fact that an investigation of rhetorical structure, its interaction with other aspects of discourse structure, and the prospects for its automatic derivation are the key motivations for building up the corpus.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The P (W |T , \u00cf\u0088) term in the lexicon component now decomposes as: n P (W |T , \u00cf\u0088) = n P (Wi|Ti, \u00cf\u0088) i=1 n \u00ef\u00a3\u00ab \u00ef\u00a3\u00b6 tions are not modeled by the standard HMM, which = n \u00ef\u00a3\u00ad n P (v|\u00cf\u0088Ti f )\u00ef\u00a3\u00b8 instead can model token-level frequency.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Finally, note that while most feature concepts are lexicalized, others, such as the suffix concept, are not.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Wu and Fung introduce an evaluation method they call nk-blind.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The first method uses a similar algorithm to that of (Yarowsky 95), with modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"encodes the one tag per word constraint and is uni form over type-level tag assignments.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Multiple features can be used for the same token.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"As expected, the vanilla HMM trained with EM performs the worst.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This intuition is born out by the experimental results.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Section 2 describes our baseline techniques for SMT adaptation, and section 3 describes the instance-weighting approach.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Unfortunately, Yarowsky's method is not well understood from a theoretical viewpoint: we would like to formalize the notion of redundancy in unlabeled data, and set up the learning task as optimization of some appropriate objective function.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The overall performance of the LTG system was outstanding, but the system consists of a sequence of many hand-coded rules and machine-learning modules.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"7 www.cis.upenn.edu\/\u00e2\u0088\u00bcpdtb\/ 8 www.eml-research.de\/english\/Research\/NLP\/ Downloads had to buy a new car.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Prague Dependency Treebank (Haji\u02c7c et al., 2001b), Danish Dependency Treebank (Kromann, 2003), and the METU Treebank of Turkish (Oflazer et al., 2003), which generally allow annotations with nonprojective dependency structures.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The observed performance gains, coupled with the simplicity of model implementation, makes it a compelling alternative to existing more complex counterparts.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"7.96 5.55 1 l...................................................................................................................................................................................................J..","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"-1 means that an NP should be ruled out as a possible antecedent, and 0 means that the knowledge source remains neutral (i.e., it has no reason to believe that they cannot be coreferent).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Each unlabeled pair (x1,i, x2,i) is represented as an edge between nodes corresponding to x1,i and X2,i in the graph.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A moment's reflection will reveal that things are not quite that simple.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For \u00c3\u0086 = 1, a new target language word is generated using the trigram language model p(eje0; e00).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The foreign language vertices (denoted by Vf) correspond to foreign trigram types, exactly as in Subramanya et al. (2010).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"(b) After they were released...","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"It also does not prune, so comparing to our pruned model would be unfair.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In turn we use two sorts of heuristics, orthogonal to one another, to prune segmentation possibilities based on lexical and grammatical constraints.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"As the name implies, space is O(m) and linear in the number of entries.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The work of the first author was supported by the Lynn and William Frankel Center for Computer Sciences.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"For example, the MCTAG shown in Figure 7 generates trees of the form shown in Figure 4b.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Modifying the Berkeley parser for Arabic is straightforward.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Semantic expectations are analogous to lexical expectations except that they represent semantic classes rather than nouns.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For novel texts, no lexicon that consists simply of a list of word entries will ever be entirely satisfactory, since the list will inevitably omit many constructions that should be considered words.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"ATB CTB6 Negra WSJ Trees 23449 28278 20602 43948 Word Typess 40972 45245 51272 46348 Tokens 738654 782541 355096 1046829 Tags 32 34 499 45 Phrasal Cats 22 26 325 27 Test OOV 16.8% 22.2% 30.5% 13.2% Per Sentence Table 4: Gross statistics for several different treebanks.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The local feature groups are: Non-Contextual Feature: This feature is set to 1 for all tokens.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We also add an annotation for one-level iDafa (oneLevelIdafa) constructs since they make up more than 75% of the iDafa NPs in the ATB (Gabbard and Kulick, 2008).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The following auxiliary quantity is defined: Qe0 (e; C; j) := probability of the best partial hypothesis (ei 1; bi 1), where C = fbkjk = 1; ; ig, bi = j, ei = e and ei\u00f4\u0080\u0080\u00801 = e0.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Examples are given in Table 4.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The sequence of states needed to carry out the word reordering example in Fig.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Like verbs, maSdar takes arguments and assigns case to its objects, whereas it also demonstrates nominal characteristics by, e.g., taking determiners and heading iDafa (Fassi Fehri, 1993).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The subtree under,; is excised from 7, the tree 7' is inserted in its place and the excised subtree is inserted below the foot of y'.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Examining the word fidanzato for the \u201cNo LP\u201d and \u201cWith LP\u201d models is particularly instructive.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Smith estimates Lotus will make profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Using a treebank grammar, a data-driven lexicon, and a linguistically motivated unknown-tokens handling technique our model outperforms previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"(If the TF\/IDF score of that word is below a threshold, the phrase is discarded.)","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In IdentiFinder, there is a priority in the feature assignment, such that if one feature is used for a token, another feature lower in priority will not be used.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"It's not clear how to apply these methods in the unsupervised case, as they required cross-validation techniques: for this reason we use the simpler smoothing method shown here. input to the unsupervised algorithm is an initial, &quot;seed&quot; set of rules.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Therefore, a populated probing hash table consists of an array of buckets that contain either one entry or are empty.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Otherwise, label the training data with the combined spelling\/contextual decision list, then induce a final decision list from the labeled examples where all rules (regardless of strength) are added to the decision list.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The knowledge base then can be tested for its relation-inference capabilities on the basis of full-blown co-reference information.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"(2010), we adopt a simpler na\u00c2\u00a8\u00c4\u00b1ve Bayes strategy, where all features are emitted independently.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"MENE (Maximum Entropy Named Entity) (Borth- wick, 1999) was combined with Proteus (a hand- coded system), and came in fourth among all MUC 7 participants.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The general-language features have a slight advantage over the similarity features, and both are better than the SVM feature.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"However, if we consider precision, recall and Fmeasure on non-projective dependencies only, as shown in Table 6, some differences begin to emerge.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"However, for multinomial models like our LMs and TMs, there is a one to one correspondence between instances and features, eg the correspondence between a phrase pair (s, t) and its conditional multinomial probability p(s1t).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Due to the dramatic fiscal situation in Brandenburg she now surprisingly withdrew legislation drafted more than a year ago, and suggested to decide on it not before 2003.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A derived structure will be mapped onto a sequence zi of substrings (not necessarily contiguous in the input), and the composition operations will be mapped onto functions that can defined as follows3. f((zi,\u2022 \u2022 \u2022 , zni), (m.,\u2022 \u2022 \u2022 ,Yn3)) = (Z1, \u2022 \u2022 \u2022 , Zn3) where each z, is the concatenation of strings from z,'s and yk's.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"more frequently than is done in English.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Let H be the set of hanzi, p be the set of pinyin syllables with tone marks, and P be the set of grammatical part-of-speech labels.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"But we will show that the use of unlabeled data can drastically reduce the need for supervision.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Step 2.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The first row represents the average accuracy of the three parsers we combine.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"(2006).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In contrast, NNP (proper nouns) form a large portion of vocabulary.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The first two rows of the table are baselines.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.4 Underspecified rhetorical structure.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"For each language and setting, we report one-to-one (11) and many- to-one (m-1) accuracies.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"AdaBoost.MH can be applied to the problem using these pseudolabels in place of supervised examples.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"As long as the main evaluation metric is dependency accuracy per word, with state-of-the-art accuracy mostly below 90%, the penalty for not handling non-projective constructions is almost negligible.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Lexicon Feature of Previous and Next Token: The string of the previous token and the next token is used with the initCaps information of . If has initCaps, then a feature (initCaps, ) is set to 1.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"As Figure 1 shows, this word has no high-confidence alignment in the Italian-English bitext.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We can use the preceding linguistic and annotation insights to build a manually annotated Arabic grammar in the manner of Klein and Manning (2003).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Word frequencies are estimated by a re-estimation procedure that involves apply\u00c2\u00ad ing the segmentation algorithm presented here to a corpus of 20 million words,8 using 8 Our training corpus was drawn from a larger corpus of mixed-genre text consisting mostly of.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We used section 23 as the development set for our combining techniques, and section 22 only for final testing.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"However, using the top-level semantic classes of WordNet proved to be problematic because the class distinctions are too coarse.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For example, if FCC and Federal Communications Commission are both found in a document, then Federal has A begin set to 1, Communications has A continue set to 1, Commission has A end set to 1, and FCC has A unique set to 1.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The development of a na\u00efve Bayes classifier involves learning how much each parser should be trusted for the decisions it makes.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"On the one hand, the type-level error rate is not calibrated for the number of n-grams in the sample.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Although the tag distributions of the foreign words (Eq.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"To be short, we omit the target words e; e0 in the formulation of the search hypotheses.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We tokenize MWUs and their POS tags; this reduces the tag set size to 12.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Part of the gap between resident and virtual memory is due to the time at which data was collected.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The use of weighted transducers in particular has the attractive property that the model, as it stands, can be straightforwardly interfaced to other modules of a larger speech or natural language system: presumably one does not want to segment Chinese text for its own sake but instead with a larger purpose in mind.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"When a comparison against previous results requires additional pre-processing, we state it explicitly to allow for the reader to replicate the reported results.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Mutual information was shown to be useful in the segmentation task given that one does not have a dictionary.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"8 1 2.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"To eliminate such sequences, we define a transition probability between word classes to be equal to 1 if the sequence is admissible, and 0 otherwise.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"To lower the barrier of entrance to the competition, we provided a complete baseline MT system, along with data resources.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"This feature ft incorporates information from the smoothed graph and prunes hidden states that are inconsistent with the thresholded vector tx.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Models To assess the marginal utility of each component of the model (see Section 3), we incremen- tally increase its sophistication.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The weight on each sentence is a value in [0, 1] computed by a perceptron with Boolean features that indicate collection and genre membership.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"(2009), who also incorporate a sparsity constraint, but does via altering the model objective using posterior regularization.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"7 Big 5 is the most popular Chinese character coding standard in use in Taiwan and Hong Kong.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Applications The discovered paraphrases have multiple applications.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"These make left-to-right query patterns convenient, as the application need only provide a state and the word to append, then use the returned state to append another word, etc.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"When this metric is less than 0.5, we expect to incur more errors' than we will remove by adding those constituents to the parse.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"This remains as future work.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"However, it is robust, efficient, and easy to implement.4 To perform the maximization in (7), we used the popular L-BFGS algorithm (Liu and Nocedal, 1989), which requires gradient information.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Even when there is training data available in the domain of interest, there is often additional data from other domains that could in principle be used to improve performance.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"One is that smaller sets sometime have meaningless keywords, like \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d in the CC-domain, or \u00e2\u0080\u009ccompare\u00e2\u0080\u009d in the PC-domain.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Finally, as\u00c2\u00ad suming a simple bigram backoff model, we can derive the probability estimate for the particular unseen word i\u00c2\u00a51J1l.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We would like to thank Ryan McDonald for numerous discussions on this topic.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Along with locking and background kernel operations such as prefaulting, this explains why wall time is not one-eighth that of the single-threaded case. aLossy compression with the same weights. bLossy compression with retuned weights. the non-lossy options.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Tsarfaty (2006) was the first to demonstrate that fully automatic Hebrew parsing is feasible using the newly available 5000 sentences treebank.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The original OUT counts co(s, t) are weighted by a logistic function w\u03bb(s, t): To motivate weighting joint OUT counts as in (6), we begin with the \u201cideal\u201d objective for setting multinomial phrase probabilities 0 = {p(s|t), dst}, which is the likelihood with respect to the true IN distribution pi(s, t).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The second setting uses the news-related subcorpora for the NIST09 MT Chinese to English evaluation8 as IN, and the remaining NIST parallel Chinese\/English corpora (UN, Hong Kong Laws, and Hong Kong Hansard) as OUT.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(b) F.i'JJI!","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"8 66.4 52.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The hash variant is a reverse trie with hash tables, a more memory-efficient version of SRILM\u2019s default.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The bias of automatic methods in favor of statistical systems seems to be less pronounced on out-of-domain test data.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Consider the following sentences: (a) Jose Maria Martinez, Roberto Lisandy, and Dino Rossy, who were staying at a Tecun Uman hotel, were kidnapped by armed men who took them to an unknown place.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Though we do not directly compare state implementations, performance metrics in Table 1 indicate our overall method is faster.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"), and thosethat begin with a verb (\u00ef\u00bf\u00bd ub..i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In Section 2, we brie y review our approach to statistical machine translation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Second, we identified the 100 most frequent nouns in the training corpus and manually labeled them with semantic tags.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"For each language, we took the same number of sentences from the bitext as there are in its treebank, and trained a supervised feature-HMM.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The learned patterns are then normalized and applied to the corpus.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We retain segmentation markers\u00e2\u0080\u0094which are consistent only in the vocalized section of the treebank\u00e2\u0080\u0094to differentiate between e.g. \u00ef\u00bf\u00bd \u00e2\u0080\u009cthey\u00e2\u0080\u009d and \u00ef\u00bf\u00bd + \u00e2\u0080\u009ctheir.\u00e2\u0080\u009d Because we use the vocalized section, we must remove null pronoun markers.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"63 95.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(3)), with one term for each classifier.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The model was built with open vocabulary, modified Kneser-Ney smoothing, and default pruning settings that remove singletons of order 3 and higher.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"G1 and G2 are hanzi, we can estimate the probability of the sequence being a name as the product of: \u00e2\u0080\u00a2 the probability that a word chosen randomly from a text will be a name-p(rule 1), and \u00e2\u0080\u00a2 the probability that the name is of the form 1hanzi-family 2hanzi-given-p(rule 2), and \u00e2\u0080\u00a2 the probability that the family name is the particular hanzi F1-p(rule 6), and \u00e2\u0080\u00a2 the probability that the given name consists of the particular hanzi G1 and G2-p(rule 9) This model is essentially the one proposed in Chang et al.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"A typical situation is shown in Figure 1.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For the translation experiments, Eq. 2 is recursively evaluated.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"3.3 Evaluation Results.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The details of the transformation procedure are slightly different depending on the encoding schemes: d\u2191h let the linear head be the syntactic head). target arc must have the form wl \u2212\u2192 wm; if no target arc is found, Head is used as backoff. must have the form wl \u2212\u2192 wm and no outgoing arcs of the form wm p'\u2193 \u2212\u2192 wo; no backoff.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"(f1; ;mg n fl1g ; l) 3 (f1; ;mg n fl; l1; l2g ; l0) !","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"For example, the MCTAG shown in Figure 7 generates trees of the form shown in Figure 4b.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"5 70.1 58.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In section 4 we evaluate these transformations with respect to projectivized dependency treebanks, and in section 5 they are applied to parser output.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Specifically, for the ith word type, the set of token-level tags associated with token occurrences of this word, denoted t(i), must all take the value Ti to have nonzero mass. Thus in the context of Gibbs sampling, if we want to block sample Ti with t(i), we only need sample values for Ti and consider this setting of t(i).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Intuitively, it places more weight on OUT when less evidence from IN is available.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"On the MUC6 data, Bikel et al.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"8 1 8.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"However, this result is consistent with the results of ex\u00c2\u00ad periments discussed in Wu and Fung (1994).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This would result in better rest cost estimation and better pruning.10 In general, tighter, but well factored, integration between the decoder and language model should produce a significant speed improvement.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The table shows that the lexicon tag frequency predicated by our full model are the closest to the gold standard.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We can make several observations on the cause of errors.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Purely statistical approaches have not been very popular, and so far as we are aware earlier work by Sproat and Shih (1990) is the only published instance of such an approach.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"5 64.7 42.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For RandLM, we used the settings in the documentation: 8 bits per value and false positive probability 1 256.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Later, BerkeleyLM (Pauls and Klein, 2011) described ideas similar to ours.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Our clue is the NE instance pairs.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Since the early days of statistical NLP, researchers have observed that a part-of-speech tag distribution exhibits \u00e2\u0080\u009cone tag per discourse\u00e2\u0080\u009d sparsity \u00e2\u0080\u0094 words are likely to select a single predominant tag in a corpus, even when several tags are possible.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Nonetheless, the prospect of a network of annotated discourse resources seems particularly promising if not only a single annotation layer is used but a whole variety of them, so that a systematic search for correlations between them becomes possible, which in turn can lead to more explanatory models of discourse structure.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We focus here instead on adapting the two most important features: the language model (LM), which estimates the probability p(wIh) of a target word w following an ngram h; and the translation models (TM) p(slt) and p(t1s), which give the probability of source phrase s translating to target phrase t, and vice versa.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"8 We use head-finding rules specified by a native speaker.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The overall performance of the LTG system was outstanding, but the system consists of a sequence of many hand-coded rules and machine-learning modules.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We would like to thank Prof. Ralph Grish- man, Mr. Takaaki Hasegawa and Mr. Yusuke Shinyama for useful comments, discussion and evaluation.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"With respect to exact match, the improvement is even more noticeable, which shows quite clearly that even if non-projective dependencies are rare on the token level, they are nevertheless important for getting the global syntactic structure correct.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We did have a problem classifying some modified noun phrases where the modified phrase does not represent a qualified or restricted form of the head, like \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, as these are both represented by the keyword \u00e2\u0080\u009cchairman\u00e2\u0080\u009d.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The PROBING model is 2.4 times as fast as the fastest alternative, SRILM, and uses less memory too.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The maximum precision oracle is an upper bound on the possible gain we can achieve by parse hybridization.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The advantage is that we can recombine search hypotheses by dynamic programming.","label":0},{"source_text":"There is no global pruning.","target_text":"= p(fj je) max \u00c3\u0086;e00 j02Cnfjg np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j 0 )o: The DP equation is evaluated recursively for each hypothesis (e0; e; C; j).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"TPT has theoretically better locality because it stores ngrams near their suffixes, thereby placing reads for a single query in the same or adjacent pages.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Phrase-level granularity distinguishes our work from previous work by Matsoukas et al (2009), who weight sentences according to sub-corpus and genre membership.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Exploiting Diversity in Natural Language Processing: Combining Parsers","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"For graph propagation, the hyperparameter v was set to 2 x 10\u22126 and was not tuned.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"All commentaries have been tagged with part-of-speech information using Brants\u00e2\u0080\u0099 TnT1 tagger and the Stuttgart\/Tu\u00c2\u00a8bingen Tag Set automatic analysis was responsible for this decision.)","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Statistics for all data sets are shown in Table 2.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Mai.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"37.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(a) I f f fi * fi :1 }'l ij 1\u00c2\u00a7: {1M m m s h e n 3 m e 0 shi2 ho u4 wo 3 cai2 ne ng 2 ke4 fu 2 zh e4 ge 4 ku n4 w h a t ti m e I just be abl e ov er co m e thi s C L dif fic 'When will I be able to overcome this difficulty?'","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"5.1 Parsing Models.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Section 4.1 explained that state s is stored by applications with partial hypotheses to determine when they can be recombined.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Lattice parsing (Chappelier et al., 1999) is an alternative to a pipeline that prevents cascading errors by placing all segmentation options into the parse chart.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Formally, for a lexicon L and segments I \u00e2\u0088\u0088 L, O \u00e2\u0088\u0088\/ L, each word automaton accepts the language I\u00e2\u0088\u0097(O + I)I\u00e2\u0088\u0097.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The results described in this paper suggest a characterization of mild context-sensitivity in terms of generalized context-freeness.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"16 As one reviewer points out, one problem with the unigram model chosen here is that there is still a. tendency to pick a segmentation containing fewer words.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Nonetheless, the prospect of a network of annotated discourse resources seems particularly promising if not only a single annotation layer is used but a whole variety of them, so that a systematic search for correlations between them becomes possible, which in turn can lead to more explanatory models of discourse structure.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In Section 6 we dis\u00c2\u00ad cuss other issues relating to how higher-order language models could be incorporated into the model.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"5 70.1 58.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Nu mb er filters candidate if number doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Both parameters depend on a single hyperparameter \u00ce\u00b1.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The CoBoost algorithm described above divides the function Zco into two parts: Zco = 40 + 40.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Our approach uses word-to-word dependencies between source and target words.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Final (F): The rest of the sentence is processed monotonically taking account of the already covered positions.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"3.2 Stochastic rhetorical analysis.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: Mr. Cristiani, president of the country ...","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Better Arabic Parsing: Baselines, Evaluations, and Analysis","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This has led previous workers to adopt ad hoc linear weighting schemes (Finch and Sumita, 2008; Foster and Kuhn, 2007; L\u00a8u et al., 2007).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(3)) to be defined over unlabeled as well as labeled instances.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Since the early days of statistical NLP, researchers have observed that a part-of-speech tag distribution exhibits \u00e2\u0080\u009cone tag per discourse\u00e2\u0080\u009d sparsity \u00e2\u0080\u0094 words are likely to select a single predominant tag in a corpus, even when several tags are possible.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Another way to interpret this is that less than 5% of the correct constituents are missing from the hypotheses generated by the union of the three parsers.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Examining the word fidanzato for the \u201cNo LP\u201d and \u201cWith LP\u201d models is particularly instructive.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As we have seen, the lexicon of basic words and stems is represented as a WFST; most arcs in this WFST represent mappings between hanzi and pronunciations, and are costless.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"This leads to the best reported performance for robust non-projective parsing of Czech.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We include a list of per-category results for selected phrasal labels, POS tags, and dependencies in Table 8.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Both of these analyses are shown in Figure 4; fortunately, the correct analysis is also the one with the lowest cost, so it is this analysis that is chosen.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We refer to a segment and its assigned PoS tag as a lexeme, and so analyses are in fact sequences of lexemes.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Both implementations employ a state object, opaque to the application, that carries information from one query to the next; we discuss both further in Section 4.2.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Recent research has demonstrated that incorporating this sparsity constraint improves tagging accuracy.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Except for the left3The graphs satisfy all the well-formedness conditions given in section 2 except (possibly) connectedness.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"For each extension a new position is added to the coverage set.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Hyperparameter settings are sorted according to the median one-to-one metric over runs.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In the last example, the less restrictive IbmS word reordering leads to a better translation, although the QmS translation is still acceptable.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For each case- frame, BABAR collects the head nouns of noun phrases that were extracted by the caseframe in the training corpus.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Values in the trie are minimally sized at the bit level, improving memory consumption over trie implementations in SRILM, IRSTLM, and BerkeleyLM.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Similarly, the tokens and are tested against each list, and if found, a corresponding feature will be set to 1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Much confusion has been sown about Chinese writing by the use of the term ideograph, suggesting that hanzi somehow directly represent ideas.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Chang of Tsinghua University, Taiwan, R.O.C., for kindly providing us with the name corpora.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"An example of a fairly low-level relation is the affix relation, which holds between a stem morpheme and an affix morpheme, such as f1 -menD (PL).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This method, one instance of which we term the \"greedy algorithm\" in our evaluation of our own system in Section 5, involves starting at the beginning (or end) of the sentence, finding the longest word starting (ending) at that point, and then repeating the process starting at the next (previous) hanzi until the end (begin\u00c2\u00ad ning) of the sentence is reached.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Table 5: Effect of the beam threshold on the number of search errors (147 sentences).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Such ambiguities cause discrepancies between token boundaries (indexed as white spaces) and constituent boundaries (imposed by syntactic categories) with respect to a surface form.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The accuracy results for segmentation, tagging and parsing using our different models and our standard data split are summarized in Table 1.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"While there are other obstacles to completing this idea, we believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Manual annotation results in human in- terpretable grammars that can inform future tree- bank annotation decisions.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The cost estimate, cost(i\u00c2\u00a5JJ1l.fn is computed in the obvious way by summing the negative log probabilities of i\u00c2\u00a5JJ1l.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"In this paper, we make a simplifying assumption of one-tag-per-word.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The original OUT counts co(s, t) are weighted by a logistic function w\u03bb(s, t): To motivate weighting joint OUT counts as in (6), we begin with the \u201cideal\u201d objective for setting multinomial phrase probabilities 0 = {p(s|t), dst}, which is the likelihood with respect to the true IN distribution pi(s, t).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"If so, the CF Network reports that the anaphor and candidate may be coreferent.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In practice, we can therefore expect a trade-off such that increasing the amount of information encoded in arc labels will cause an increase in the accuracy of the inverse transformation but a decrease in the accuracy with which the parser can construct the labeled representations.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The first probability is estimated from a name count in a text database, and the rest of the probabilities are estimated from a large list of personal names.n Note that in Chang et al.'s model the p(rule 9) is estimated as the product of the probability of finding G 1 in the first position of a two-hanzi given name and the probability of finding G2 in the second position of a two-hanzi given name, and we use essentially the same estimate here, with some modifications as described later on.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The restriction can be expressed in terms of the number of uncovered source sentence positions to the left of the rightmost position m in the coverage set.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"One knowledge source, called WordSemCFSem, is analogous to CFLex: it checks whether the anaphor and candidate antecedent are substitutable for one another, but based on their semantic classes instead of the words themselves.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Wu and Fung introduce an evaluation method they call nk-blind.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"One implementation issue deserves some elaboration.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"To lower the barrier of entrance to the competition, we provided a complete baseline MT system, along with data resources.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Same for . In the case where the next token is a hyphen, then is also used as a feature: (init- Caps, ) is set to 1.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We thank Felix Hageloh (Hageloh, 2006) for providing us with this version. proposed in (Tsarfaty, 2006).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"gaolxing4 'happy' => F.i'JF.i'J Jl!","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"mein 5.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"\u00e2\u0080\u00a2 Bridging links: the annotator is asked to specify the type as part-whole, cause-effect (e.g., She had an accident.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We also mark all nodes that dominate an SVO configuration (containsSVO).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Label propagation can then be used to transfer the labels to the peripheral foreign vertices (i.e. the ones adjacent to the English vertices) first, and then among all of the foreign vertices (\u00a74).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Therefore, a populated probing hash table consists of an array of buckets that contain either one entry or are empty.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"A graphical depiction of our model as well as a summary of random variables and parameters can be found in Figure 1.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Past work however, has typically associ n = n P (Ti)P (Wi|Ti) = i=1 1 n K n ated these features with token occurrences, typically in an HMM.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"To see this, note thai the first two terms in the above equation correspond to the function that AdaBoost attempts to minimize in the standard supervised setting (Equ.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"35 76.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The details of the transformation procedure are slightly different depending on the encoding schemes: d\u2191h let the linear head be the syntactic head). target arc must have the form wl \u2212\u2192 wm; if no target arc is found, Head is used as backoff. must have the form wl \u2212\u2192 wm and no outgoing arcs of the form wm p'\u2193 \u2212\u2192 wo; no backoff.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Since the early days of statistical NLP, researchers have observed that a part-of-speech tag distribution exhibits \u00e2\u0080\u009cone tag per discourse\u00e2\u0080\u009d sparsity \u00e2\u0080\u0094 words are likely to select a single predominant tag in a corpus, even when several tags are possible.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Some of the operations will be constant functions, corresponding to elementary structures, and will be written as f () = zi), where each z, is a constant, the string of terminal symbols al an,,,.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"A more forceful approach for encoding sparsity is posterior regularization, which constrains the posterior to have a small number of expected tag assignments (Grac\u00c2\u00b8a et al., 2009).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Clearly the percentage of productively formed words is quite small (for this particular corpus), meaning that dictionary entries are covering most of the 15 GR is .73 or 96%..","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"keywords Step 3 Sets of phrases based on keywords Step 4 Links between sets of phrases All the contexts collected for a given domain are gathered in a bag and the TF\/ITF scores are calculated for all the words except stopwords in the bag.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Turning now to (1), we have the similar problem that splitting.into.ma3 'horse' andlu4 'way' is more costly than retaining this as one word .ma3lu4 'road.'","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Note that it is in precision that our over\u00c2\u00ad all performance would appear to be poorer than the reported performance of Chang et al., yet based on their published examples, our system appears to be doing better precisionwise.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"31 75.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The terms on the right-hand-side denote the type-level and token-level probability terms respectively.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.1 Reliable Case Resolutions.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"For example, the two NEs \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d and \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d have the following contexts.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Evalb is a Java re-implementation of the standard labeled precision\/recall metric.12 The ATB gives all punctuation a single tag.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Not all the layers have been produced for all the texts yet.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Table 4 shows the results.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Every parse \u03c0 selects a specific morphological segmentation (l1...lk) (a path through the lattice).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In particular, it may not be possible to learn functions fi (x f2(x2,t) for i = m + 1...n: either because there is some noise in the data, or because it is just not realistic to expect to learn perfect classifiers given the features used for representation.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In the last example, the less restrictive IbmS word reordering leads to a better translation, although the QmS translation is still acceptable.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"There is a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099 of ten commentaries, for which the range of information (except for syntax) has been completed; the remaining data has been annotated to different degrees, as explained below.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"For the translation model Pr(fJ 1 jeI 1), we go on the assumption that each source word is aligned to exactly one target word.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"2.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Model components cascade, so the row corresponding to +FEATS also includes the PRIOR component (see Section 3).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"During testing, it is possible that the classifier produces a sequence of inadmissible classes (e.g., person begin followed by location unique).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"41.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The first row represents the average accuracy of the three parsers we combine.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This paper describes the several performance techniques used and presents benchmarks against alternative implementations.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"At present, the \u00e2\u0080\u0098Potsdam Commentary Corpus\u00e2\u0080\u0099 (henceforth \u00e2\u0080\u0098PCC\u00e2\u0080\u0099 for short) consists of 170 commentaries from Ma\u00c2\u00a8rkische Allgemeine Zeitung, a German regional daily.","label":1},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The type of alignment we have considered so far requires the same length for source and target sentence, i.e. I = J. Evidently, this is an unrealistic assumption, therefore we extend the concept of inverted alignments as follows: When adding a new position to the coverage set C, we might generate either \u00c3\u0086 = 0 or \u00c3\u0086 = 1 new target words.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Table 4 shows translation results for the three approaches.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We currently simulate this by crafting a WCFG and feeding it to BitPar.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Links can be of two different kinds: anaphoric or bridging (definite noun phrases picking up an antecedent via world-knowledge).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Machine Translation In this case my colleague can not visit on I n d i e s e m F a l l ka nn m e i n K o l l e g e a m the v i e r t e n M a i n i c h t b e s u c h e n S i e you fourth of May Figure 1: Reordering for the German verbgroup.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Among these are words derived by various productive processes, including: 1.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Many hanzi have more than one pronunciation, where the correct.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"(c) After they blindfolded the men...","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"(c) Coordination ambiguity is shown in dependency scores by e.g., \u00e2\u0088\u0097SSS R) and \u00e2\u0088\u0097NP NP NP R).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The implication of this ambiguity for a parser is that the yield of syntactic trees no longer consists of spacedelimited tokens, and the expected number of leaves in the syntactic analysis in not known in advance.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The method halves the error rate in comparison to a method using the labeled examples alone.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"At each iteration the algorithm increases the number of rules, while maintaining a high level of agreement between the spelling and contextual decision lists.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"It also uses less memory, with 8 bytes of overhead per entry (we store 16-byte entries with m = 1.5); linked list implementations hash set and unordered require at least 8 bytes per entry for pointers.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"A procedural definition to restrict1In the approach described in (Berger et al., 1996), a mor phological analysis is carried out and word morphemes rather than full-form words are used during the search.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Applications such as machine translation use language model probability as a feature to assist in choosing between hypotheses.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We realize the importance of paraphrase; however, the major obstacle is the construction of paraphrase knowledge.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"A step of an ATM consists of reading a symbol from each tape and optionally moving each head to the left or right one tape cell.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Overall, it gives improvements ranging from 1.1% for German to 14.7% for Italian, for an average improvement of 8.3% over the unsupervised feature-HMM model.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"\u00ce\u00b8 has a belief value of 1.0, indicating complete certainty that the correct hypothesis is included in the set, and a plausibility value of 1.0, indicating that there is no evidence for competing hypotheses.5 As evidence is collected and the likely hypotheses are whittled down, belief is redistributed to subsets of \u00ce\u00b8.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Table 2 Similarity matrix for segmentation judgments.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"This remains as future work.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"On the surface, our model may seem as a special case of Cohen and Smith in which \u03b1 = 0.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"A contextual rule considers words surrounding the string in the sentence in which it appears (e.g., a rule that any proper name modified by an appositive whose head is president is a person).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Two aspects of the corpus have been presented in previous papers ((Re- itter, Stede 2003) on underspecified rhetorical structure; (Stede 2003) on the perspective of knowledge-based summarization).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Since our destructor is an efficient call to munmap, bypassing the destructor favors only other packages.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"With the additional assumption that (s, t) can be restricted to the support of co(s, t), this is equivalent to a \u201cflat\u201d alternative to (6) in which each non-zero co(s, t) is set to one.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We were intentionally lenient with our baselines: bilingual information by projecting POS tags directly across alignments in the parallel data.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The portions of information in the large window can be individually clicked visible or invisible; here we have chosen to see (from top to bottom) \u00e2\u0080\u00a2 the full text, \u00e2\u0080\u00a2 the annotation values for the activated annotation set (co-reference), \u00e2\u0080\u00a2 the actual annotation tiers, and \u00e2\u0080\u00a2 the portion of text currently \u00e2\u0080\u0098in focus\u00e2\u0080\u0099 (which also appears underlined in the full text).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"By contrast, BerkeleyLM\u2019s hash and compressed variants will return incorrect results based on an n \u22121-gram.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In our implementation, we make perhaps the simplest choice of weak hypothesis.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We maintain a separate array for each length n containing all n-gram entries sorted in suffix order.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"When translating the sentence monotonically from left to right, the translation of the German finite verb 'kann', which is the left verbal brace in this case, is postponed until the German noun phrase 'mein Kollege' is translated, which is the subject of the sentence.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Unlabeled examples in the named-entity classification problem can reduce the need for supervision to a handful of seed rules.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We assume that M is in an existential state qA, with integers i1 and i2 representing zi in the (2i \u2014 1)th and 22th work tape, for 1 < i < k. For each rule p : A fp(B, C) such that fp is mapped onto the function fp defined by the following rule. jp((xi,.. \u2022 ,rnt), (1ii, \u2022 \u2022 \u2022 \u2022 Yn3))= (Zi , \u2022 \u2022 \u2022 , Zk) M breaks xi , zk into substrings xi, , xn, and yi,...,y&quot; conforming to the definition of fp.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"On the other hand, we can expect Head+Path to be the most useful representation for reconstructing the underlying non-projective dependency graph.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"On several languages, we report performance exceeding that of state-of-the art systems.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We therefore used the arithmetic mean of each interjudge precision-recall pair as a single measure of interjudge similarity.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Information from a sentence is sometimes insufficient to classify a name correctly.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This paper, however, provides a comprehensive overview of the data collection effort and its current state.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The number of NE instance pairs used in their experiment is less than half of our method.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"A corpus of German newspaper commentaries has been assembled and annotated with different information (and currently, to different degrees): part-of-speech, syntax, rhetorical structure, connectives, co-reference, and information structure.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The most frequent NE category pairs are \u00e2\u0080\u009cPerson - Person (209,236), followed by \u00e2\u0080\u009cCountry - Coun- try\u00e2\u0080\u009d (95,123) and \u00e2\u0080\u009cPerson - Country\u00e2\u0080\u009d (75,509).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"As expected, the vanilla HMM trained with EM performs the worst.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The first stage identifies a keyword in each phrase and joins phrases with the same keyword into sets.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Note that in our construction arcs can never cross token boundaries.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Thus we have some confidence that our own performance is at least as good as that of Chang et al.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Hence, the resulting trees will no longer have two branches of equal size.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"C is the union of the sets of constituents suggested by the parsers. r(c) is a binary function returning t (for true) precisely when the constituent c E C should be included in the hypothesis.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The first probability is estimated from a name count in a text database, and the rest of the probabilities are estimated from a large list of personal names.n Note that in Chang et al.'s model the p(rule 9) is estimated as the product of the probability of finding G 1 in the first position of a two-hanzi given name and the probability of finding G2 in the second position of a two-hanzi given name, and we use essentially the same estimate here, with some modifications as described later on.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"0 57.2 43.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Finally, we make some improvements to baseline approaches.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"As is standard, we report the greedy one-to-one (Haghighi and Klein, 2006) and the many-to-one token-level accuracy obtained from mapping model states to gold POS tags.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The record for wn1 stores the offset at which its extensions begin.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"On several languages, we report performance exceeding that of more complex state-of-the art systems.1","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"We respond to this on the one hand with a format for its underspecification (see 2.4) and on the other hand with an additional level of annotation that attends only to connectives and their scopes (see 2.5), which is intended as an intermediate step on the long road towards a systematic and objective treatment of rhetorical structure.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"First, a non-anaphoric NP classifier identifies definite noun phrases that are existential, using both syntactic rules and our learned existential NP recognizer (Bean and Riloff, 1999), and removes them from the resolution process.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"At each iteration the algorithm increases the number of rules, while maintaining a high level of agreement between the spelling and contextual decision lists.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Email: cls@bell-labs.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The zone to which a token belongs is used as a feature.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Morphological segmentation decisions in our model are delegated to a lexeme-based PCFG and we show that using a simple treebank grammar, a data-driven lexicon, and a linguistically motivated unknown-tokens handling our model outperforms (Tsarfaty, 2006) and (Cohen and Smith, 2007) on the joint task and achieves state-of-the-art results on a par with current respective standalone models.2","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"5.2 Setup.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Juri Ganitkevitch answered questions about Joshua.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"01 75.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"It was also proposed to allow annotators to skip sentences that they are unable to judge.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"att.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This range is collapsed to a number of buckets, typically by taking the hash modulo the number of buckets.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Linear probing hash tables must have more buckets than entries, or else an empty bucket will never be found.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Finally, the concatenated 5 * 20% output is used to train the reference resolution component.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"As we have said, parse quality decreases with sentence length.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"This design does not guarantee \u00e2\u0080\u009cstructural zeros,\u00e2\u0080\u009d but biases towards sparsity.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"They are: 5We are grateful to an anonymous reviewer for pointing this out.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Section 2.1 describes how BABAR generates training examples to use in the learning process.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The implementation is In a similar effort, (G\u00c2\u00a8otze 2003) developed a proposal for the theory-neutral annotation of information structure (IS) \u00e2\u0080\u0094 a notoriously difficult area with plenty of conflicting and overlapping terminological conceptions.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"However, lazy mapping is generally slow because queries against uncached pages must wait for the disk.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We pick the parse that is most similar to the other parses by choosing the one with the highest sum of pairwise similarities.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"To evaluate the performance on the segmentation task, we report SEG, the standard harmonic means for segmentation Precision and Recall F1 (as defined in Bar-Haim et al. (2005); Tsarfaty (2006)) as well as the segmentation accuracy SEGTok measure indicating the percentage of input tokens assigned the correct exact segmentation (as reported by Cohen and Smith (2007)).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The reference medicine for Silapo is EPREX\/ERYPO, which contains epoetin alfa.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Note that we check for , the word preceding the consecutive sequence of initCaps tokens, since person prefixes like Mr., Dr., etc are not part of person names, whereas corporate suffixes like Corp., Inc., etc are part of corporate names.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Second, comparisons of different methods are not meaningful unless one can eval\u00c2\u00ad uate them on the same corpus.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.2.1 The Caseframe Representation Information extraction (IE) systems use extraction patterns to identify noun phrases that play a specific role in 1 Our implementation only resolves NPs that occur in the same document, but in retrospect, one could probably resolve instances of the same existential NP in different documents too.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Various verbal (e.g., \u00ef\u00bf\u00bd, .::) and adjectival.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"For these models we limit the options provided for OOV words by not considering the entire token as a valid segmentation in case at least some prefix segmentation exists.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"To date we have not done a separate evaluation of foreign-name recognition.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Formally, for a lexicon L and segments I \u00e2\u0088\u0088 L, O \u00e2\u0088\u0088\/ L, each word automaton accepts the language I\u00e2\u0088\u0097(O + I)I\u00e2\u0088\u0097.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"52 15.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Model Overview The model starts by generating a tag assignment T for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"One class comprises words derived by productive morphologi\u00c2\u00ad cal processes, such as plural noun formation using the suffix ir, menD.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"First, from a large corpus, we extract all the NE instance pairs.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Experiments in two domains showed that the contextual role knowledge improved coreference performance, especially on pronouns.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"It would have therefore also been possible to use the integer programming (IP) based approach of Ravi and Knight (2009) instead of the feature-HMM for POS induction on the foreign side.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We also mark all tags that dominate a word with the feminine ending :: taa mar buuTa (markFeminine).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The implementation is In a similar effort, (G\u00c2\u00a8otze 2003) developed a proposal for the theory-neutral annotation of information structure (IS) \u00e2\u0080\u0094 a notoriously difficult area with plenty of conflicting and overlapping terminological conceptions.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, there is a strong relationship between ni1s and the number of hanzi in the class.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For example, if FCC and Federal Communications Commission are both found in a document, then Federal has A begin set to 1, Communications has A continue set to 1, Commission has A end set to 1, and FCC has A unique set to 1.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Members of LCFRS whose operations have this property can be translated into the ILFP notation (Rounds, 1985).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":".., Tn ) T W \u00cf\u0086 E w : Token word seqs (obs) t : Token tag assigns (det by T ) PARAMETERS \u00cf\u0088 : Lexicon parameters \u00ce\u00b8 : Token word emission parameters \u00cf\u0086 : Token tag transition parameters \u00cf\u0086 \u00cf\u0086 t1 t2 \u00ce\u00b8 \u00ce\u00b8 w1 w2 K \u00cf\u0086 T tm O K \u00ce\u00b8 E wN m N N Figure 1: Graphical depiction of our model and summary of latent variables and parameters.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In figure 4, reverse relations are indicated by `*\u00e2\u0080\u0099 next to the frequency.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Many researchers have developed coreference resolvers, so we will only discuss the methods that are most closely related to BABAR.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We evaluated BABAR on two domains: terrorism and natural disasters.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"As with lexical expections, the semantic classes of co-referring expressions are 4 They may not be perfectly substitutable, for example one NP may be more specific (e.g., \u00e2\u0080\u009che\u00e2\u0080\u009d vs. \u00e2\u0080\u009cJohn F. Kennedy\u00e2\u0080\u009d).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Next, for each pair of NE categories, we collect all the contexts and find the keywords which are topical for that NE category pair.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"3.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The set of n-grams appearing in a model is sparse, and we want to efficiently find their associated probabilities and backoff penalties.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The difference is due to more careful S-NOM NP NP NP VP VBG :: b NP restoring NP ADJP NN :: b NP NN NP NP ADJP DTJJ ADJP DTJJ NN :: b NP NP NP ADJP ADJP DTJJ J ..i NN :: b NP NP NP ADJP ADJP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ DTJJ J ..i _;\u00ef\u00bf\u00bd PRP J ..i NN _;\u00ef\u00bf\u00bd NP PRP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ J ..i role its constructive effective (b) Stanford (c) Berkeley (d) Bik el (a) Reference Figure 4: The constituent Restoring of its constructive and effective role parsed by the three different models (gold segmentation).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Finally, note that while most feature concepts are lexicalized, others, such as the suffix concept, are not.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Unfortunately, modifying the model to account for these kind of dependencies is not at all straightforward.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For example, we might have VP \u00e2\u0086\u0092 VB NP PP, where the NP is the subject.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As a first step towards modeling transliterated names, we have collected all hanzi occurring more than once in the roughly 750 foreign names in our dictionary, and we estimate the probabil\u00c2\u00ad ity of occurrence of each hanzi in a transliteration (pTN(hanzi;)) using the maximum likelihood estimate.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Within the RST \u00e2\u0080\u009cuser community\u00e2\u0080\u009d there has also been discussion whether two levels of discourse structure should not be systematically distinguished (intentional versus informational).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"7 Big 5 is the most popular Chinese character coding standard in use in Taiwan and Hong Kong.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The approach recursively evaluates a quantity Q(C; j), where C is the set of already visited cities and sj is the last visited city.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"This algorithm can be applied to statistical machine translation.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In our implementation, we make perhaps the simplest choice of weak hypothesis.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"07 80.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In MUC6 and MUC7, the named entity task is defined as finding the following classes of names: person, organization, location, date, time, money, and percent (Chinchor, 1998; Sundheim, 1995) Machine learning systems in MUC6 and MUC 7 achieved accuracy comparable to rule-based systems on the named entity task.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"It was also proposed to allow annotators to skip sentences that they are unable to judge.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"To measure the contribution of each modification, a third, intermediate algorithm, Yarowsky-cautious was also tested.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In section 2 we introduce the graph transformation techniques used to projectivize and deprojectivize dependency graphs, and in section 3 we describe the data-driven dependency parser that is the core of our system.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We implement two data structures: PROBING, designed for speed, and TRIE, optimized for memory.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Step 4.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"This assumption, however, is not inherent to type-based tagging models.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Can we do . QmS: Yes, wonderful.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"32 81.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Alternatively, h can be thought of as defining a decision list of rules x y ranked by their &quot;strength&quot; h(x, y).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"There any many techniques for improving language model speed and reducing memory consumption.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The belief value that would have been assigned to the intersection of these sets is .60*.70=.42, but this belief has nowhere to go because the null set is not permissible in the model.7 So this probability mass (.42) has to be redistributed.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The natural baseline approach is to concatenate data from IN and OUT.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The improvement is due to the cost of bit-level reads and avoiding reads that may fall in different virtual memory pages.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For example, in Northern dialects (such as Beijing), a full tone (1, 2, 3, or 4) is changed to a neutral tone (0) in the final syllable of many words: Jll donglgual 'winter melon' is often pronounced donglguaO.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"First, we use a novel graph-based framework for projecting syntactic information across language boundaries.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Comparison with state-of-the-art taggers For comparison we consider two unsupervised tag- gers: the HMM with log-linear features of Berg- Kirkpatrick et al.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"selected; and that recall is defined to be the number of correct hits divided by the number of items that should have been selected.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Eight out of the thirteen errors in the high frequency phrases in the CC-domain are the phrases in \u00e2\u0080\u009cagree\u00e2\u0080\u009d.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"971,746 sentences of New York Times text were parsed using the parser of (Collins 96).1 Word sequences that met the following criteria were then extracted as named entity examples: whose head is a singular noun (tagged NN).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We can obtain a letter equivalent CFL defined by a CFG in which the for each rule as above, we have the production A \u2014* A1 Anup where tk (up) = cp.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For even larger models, storing counts (Talbot and Osborne, 2007; Pauls and Klein, 2011; Guthrie and Hepple, 2010) is a possibility.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"92 77.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Specifically, the limit n starts at 5 and increases by 5 at each iteration.)","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Therefore, we only score guess\/gold pairs with identical character yields, a condition that allows us to measure parsing, tagging, and segmentation accuracy by ignoring whitespace.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"To prevent this we &quot;smooth&quot; the confidence by adding a small value, e, to both W+ and W_, giving at = Plugging the value of at from Equ.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Keyword detection error Even if a keyword consists of a single word, there are words which are not desirable as keywords for a domain.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We used these three parsers to explore parser combination techniques.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"During training, we treat as observed the language word types W as well as the token-level corpus w. We utilize Gibbs sampling to approximate our collapsed model posterior: P (T ,t|W , w, \u00ce\u00b1, \u00ce\u00b2) \u00e2\u0088\u009d P (T , t, W , w|\u00ce\u00b1, \u00ce\u00b2) 0.7 0.6 0.5 0.4 0.3 English Danish Dutch Germany Portuguese Spanish Swedish = P (T , t, W , w, \u00cf\u0088, \u00ce\u00b8, \u00cf\u0086, w|\u00ce\u00b1, \u00ce\u00b2)d\u00cf\u0088d\u00ce\u00b8d\u00cf\u0086 Note that given tag assignments T , there is only one setting of token-level tags t which has mass in the above posterior.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The increase is generally higher for PDT than for DDT, which indicates a greater diversity in non-projective constructions.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The less favored reading may be selected in certain contexts, however; in the case of , for example, the nominal reading jiang4 will be selected if there is morphological information, such as a following plural affix ir, menD that renders the nominal reading likely, as we shall see in Section 4.3.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"3 54.4 33.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The most frequently occurring last words of organization names in cslist are compiled into a list of corporate suffixes, Corporate-Suffix- List.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Not all the layers have been produced for all the texts yet.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"It is. based on the traditional character set rather than the simplified character set used in Singapore and Mainland China.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank, leaving only sections 22 and 23 completely untouched during the development of any of the parsers.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The table shows that the lexicon tag frequency predicated by our full model are the closest to the gold standard.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"0 Figure 5 An example of affixation: the plural affix.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Here, annotation proceeds in two phases: first, the domains and the units of IS are marked as such.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Variants of alif are inconsistently used in Arabic texts.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"(Thus the domain of the dev and test corpora matches IN.)","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In this section, we will explain the algorithm step by step with examples.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The intuition here is that the role of a discourse marker can usually be de 9 Both the corpus split and pre-processing code are avail-.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For displaying and querying the annoated text, we make use of the Annis Linguistic Database developed in our group for a large research effort (\u00e2\u0080\u0098Sonderforschungsbereich\u00e2\u0080\u0099) revolving around 9 2.7 Information structure.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"67 95.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold, 30.","label":1},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Examining the word fidanzato for the \u201cNo LP\u201d and \u201cWith LP\u201d models is particularly instructive.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Having explained the various layers of annotation in PCC, we now turn to the question what all this might be good for.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Each knowledge source then assigns a probability estimate to each candidate, which represents its belief that the candidate is the antecedent for the anaphor.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This is not ideal for some applications, however.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"(1) CEO of McCann . . .","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"A generative model was applied (similar to naive Bayes) with the three labels as hidden vanables on unlabeled examples, and observed variables on (seed) labeled examples.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"First, it directly encodes linguistic intuitions about POS tag assignments: the model structure reflects the one-tag-per-word property, and a type- level tag prior captures the skew on tag assignments (e.g., there are fewer unique determiners than unique nouns).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Part-of-speech (POS) tag distributions are known to exhibit sparsity \u00e2\u0080\u0094 a word is likely to take a single predominant tag in a corpus.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Its correct antecedent is \u00e2\u0080\u009ca revolver\u00e2\u0080\u009d, which is extracted by the caseframe \u00e2\u0080\u009ckilled with <NP>\u00e2\u0080\u009d.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In our grammar, features are realized as annotations to basic category labels.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In addition, as the CRF and PCFG look at similar sorts of information from within two inherently different models, they are far from independent and optimizing their product is meaningless.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The less favored reading may be selected in certain contexts, however; in the case of , for example, the nominal reading jiang4 will be selected if there is morphological information, such as a following plural affix ir, menD that renders the nominal reading likely, as we shall see in Section 4.3.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(student+plural) 'students,' which is derived by the affixation of the plural affix f, menD to the nounxue2shengl.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(Blum and Mitchell 98) give an example that illustrates just how powerful the second constraint can be.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses a named entity recognizer to identify proper names that refer to people and companies.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In this case, Maury Cooper is extracted.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"11 www.ling.unipotsdam.de\/sfb\/projekt a3.php 12 This step was carried out in the course of the diploma thesis work of David Reitter (2003), which de serves special mention here.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The idea is to have a pipeline of shallow-analysis modules (tagging, chunk- ing, discourse parsing based on connectives) and map the resulting underspecified rhetorical tree (see Section 2.4) into a knowledge base that may contain domain and world knowledge for enriching the representation, e.g., to resolve references that cannot be handled by shallow methods, or to hypothesize coherence relations.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(8) can now be rewritten5 as which is of the same form as the function Zt used in AdaBoost.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"All commentaries have been annotated with rhetorical structure, using RSTTool4 and the definitions of discourse relations provided by Rhetorical Structure Theory (Mann, Thompson 1988).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The Penn Arabic Treebank (ATB) syntactic guidelines (Maamouri et al., 2004) were purposefully borrowed without major modification from English (Marcus et al., 1993).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The second stage links sets which involve the same pairs of individual NEs.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(For some recent corpus-based work on Chinese abbreviations, see Huang, Ahrens, and Chen [1993].)","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"(Bar-Haim et al., 2007; Habash and Rambow, 2005)) and probabilities are assigned to different analyses in accordance with the likelihood of their tags (e.g., \u201cfmnh is 30% likely to be tagged NN and 70% likely to be tagged REL+VB\u201d).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This paper, however, provides a comprehensive overview of the data collection effort and its current state.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"mark- ContainsVerb is especially effective for distinguishing root S nodes of equational sentences.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The learned information was recycled back into the resolver to improve its performance.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"As we have noted in Section 2, the general semantic class to which a hanzi belongs is often predictable from its semantic radical.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Ideally we\u00e2\u0080\u0099d like to know the thematic role of each extracted noun phrase, but AutoSlog does not generate thematic roles.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We evaluated the results based on two metrics.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Table 1 shows results of the benchmark.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Adding the isolated constituents to our hypothesis parse could increase our expected recall, but in the cases we investigated it would invariably hurt our precision more than we would gain on recall.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Let s = a + b.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The correct ambiguity resolution of the syntactic level therefore helps to resolve the morphological one, and vice versa.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We describe a novel approach for inducing unsupervised part-of-speech taggers for languages that have no labeled training data, but have translated text in a resource-rich language.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"As is standard, we use a fixed constant K for the number of tagging states.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"SSER: subjective sentence error rate: For a more detailed analysis, the translations are judged by a human test person.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"If there are too many distinct states, the decoder prunes low-scoring partial hypotheses, possibly leading to a search error.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"During training, we treat as observed the language word types W as well as the token-level corpus w. We utilize Gibbs sampling to approximate our collapsed model posterior: P (T ,t|W , w, \u00ce\u00b1, \u00ce\u00b2) \u00e2\u0088\u009d P (T , t, W , w|\u00ce\u00b1, \u00ce\u00b2) 0.7 0.6 0.5 0.4 0.3 English Danish Dutch Germany Portuguese Spanish Swedish = P (T , t, W , w, \u00cf\u0088, \u00ce\u00b8, \u00cf\u0086, w|\u00ce\u00b1, \u00ce\u00b2)d\u00cf\u0088d\u00ce\u00b8d\u00cf\u0086 Note that given tag assignments T , there is only one setting of token-level tags t which has mass in the above posterior.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"'Malaysia.'","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A typical situation is shown in Figure 1.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Our analysis and comparison focuses primarily on the one-to-one accuracy since it is a stricter metric than many-to-one accuracy, but also report many-to-one for completeness.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We are currently exploring other methods that employ similar ideas and their formal properties.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"But it conflates the coordinating and discourse separator functions of wa (<..4.b \u00ef\u00bf\u00bd \u00ef\u00bf\u00bd) into one analysis: conjunction(Table 3).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Since the inclusion of out-ofdomain test data was a very late decision, the participants were not informed of this.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We will show that by giving the first model some global features, MENERGI outperforms Borthwick' s reference resolution classifier.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Figure 2 shows examples of lexical expectations that were learned for both domains.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"There are two possible reasons: (1) the knowledge sources are resolving different cases of anaphora, and (2) the knowledge sources provide multiple pieces of evidence in support of (or against) a candidate, thereby acting synergistically to push the DempsterShafer model over the belief threshold in favor of a single candidate.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"wo rd => na m e 2.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We do not show the numbers for the Bayes models in Table 2 because the parameters involved were established using this set.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Furthermore, we do not connect the English vertices to each other, but only to foreign language vertices.4 The graph vertices are extracted from the different sides of a parallel corpus (De, Df) and an additional unlabeled monolingual foreign corpus Ff, which will be used later for training.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The first step in the learning process is to generate training examples consisting of anaphor\/antecedent resolutions.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We would also like to thank Amarnag Subramanya for helping us with the implementation of label propagation and Shankar Kumar for access to the parallel data.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Case and Zone of and : Similarly, if (or ) is initCaps, a feature (initCaps, zone) (or (initCaps, zone) ) is set to 1, etc. Token Information: This group consists of 10 features based on the string , as listed in Table 1.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"collected too.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The evidence may not say anything about whether A is more likely than B, only that C and D are not likely.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Model components cascade, so the row corresponding to +FEATS also includes the PRIOR component (see Section 3).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Other work includes transferring latent topic distributions from source to target language for LM adaptation, (Tam et al., 2007) and adapting features at the sentence level to different categories of sentence (Finch and Sumita, 2008).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"In contrast to these approaches, our method directly incorporates these constraints into the structure of the model.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(b) F.i'JJI!","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We present several variations for the lexical component P (T , W |\u00cf\u0088), each adding more complex pa rameterizations.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The second modification is more important, and is discussed in the next section.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The matching sentence pairs are then added to the IN corpus, and the system is re-trained.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Gather phrases using keywords Next, we select a keyword for each phrase \u00e2\u0080\u0093 the top-ranked word based on the TF\/IDF metric.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses a DempsterShafer decision model (Stefik, 1995) to combine the evidence provided by the knowledge sources.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We received submissions from 14 groups from 11 institutions, as listed in Figure 2.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Since Dane\u00cb\u0087s\u00e2\u0080\u0099 proposals of \u00e2\u0080\u0098thematic development patterns\u00e2\u0080\u0099, a few suggestions have been made as to the existence of a level of discourse structure that would predict the information structure of sentences within texts.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"43 80.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The following recursive equation is evaluated: Qe0 (e; S; C; j) = (2) = p(fj je) max \u00c3\u0086;e00 np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) max (S0;j0) (S0 ;Cnfjg;j0)!(S;C;j) j02Cnfjg Qe00 (e0; S0;C n fjg; j0)o: The search ends in the hypotheses (I; f1; ; Jg; j).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"5 We choose these two metrics over the Variation Information measure due to the deficiencies discussed in Gao and Johnson (2008).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In our second model GTvpi we also distinguished finite and non-finite verbs and VPs as 10Lattice parsing can be performed by special initialization of the chart in a CKY parser (Chappelier et al., 1999).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The first issue relates to the completeness of the base lexicon.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"However, such an approach requires a number of hand-coded systems, which may not be available in languages other than English.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Among these are words derived by various productive processes, including: 1.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For the seen word ir, 'gen\u00c2\u00ad erals,' there is an c:NC transduction from to the node preceding ir,; this arc has cost cost( f,) - cost(unseen(f,)), so that the cost of the whole path is the desired cost( f,).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"A corollary of the result of Section 4.3 is that polynomial time recognition of MCTAG's is possible.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"6.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Contains digits.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The developers suggested some changes, such as building the model from scratch with IRSTLM, but these did not resolve the problem.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The question of what soft function to pick, and how to design' algorithms which optimize it, is an open question, but appears to be a promising way of looking at the problem.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"This is in contrast to dependency treebanks, e.g.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The traveling salesman problem is an optimization problem which is defined as follows: given are a set of cities S = s1; ; sn and for each pair of cities si; sj the cost dij > 0 for traveling from city si to city sj . We are looking for the shortest tour visiting all cities exactly once while starting and ending in city s1.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"An anti-greedy algorithm, AG: instead of the longest match, take the.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Clearly, explicitly modeling such a powerful constraint on tagging assignment has a potential to significantly improve the accuracy of an unsupervised part-of-speech tagger learned without a tagging dictionary.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"2.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The corpora for both settings are summarized in table 1.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Two lists, Corporate-Suffix-List (for corporate suffixes) and Person-Prefix-List (for person prefixes), are collected from the training data.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For example, a person\u00e2\u0080\u0099s full name will match with just their last name (e.g., \u00e2\u0080\u009cGeorge Bush\u00e2\u0080\u009d and \u00e2\u0080\u009cBush\u00e2\u0080\u009d), and a company name will match with and without a corporate suffix (e.g., \u00e2\u0080\u009cIBM Corp.\u00e2\u0080\u009d and \u00e2\u0080\u009cIBM\u00e2\u0080\u009d).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"4.1 The Task and the Corpus.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"A morphological analyzer M : W\u2014* L is a function mapping sentences in Hebrew (W E W) to their corresponding lattices (M(W) = L E L).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We would like to thank Eugene Charniak, Michael Collins, and Adwait Ratnaparkhi for enabling all of this research by providing us with their parsers and helpful comments.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"3.1 Word ReOrdering with Verbgroup.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In this example there are four \"input characters,\" A, B, C and D, and these map respectively to four \"pronunciations\" a, b, c and d. Furthermore, there are four \"words\" represented in the dictionary.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"As already pointed out in Section 2.4, current theories diverge not only on the number and definition of relations but also on apects of structure, i.e., whether a tree is sufficient as a representational device or general graphs are required (and if so, whether any restrictions can be placed on these graph\u00e2\u0080\u0099s structures \u00e2\u0080\u0094 cf.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In total 13,976 phrases are assigned to sets of phrases, and the accuracy on our evaluation data ranges from 65 to 99%, depending on the domain and the size of the sets.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Keyword detection error Even if a keyword consists of a single word, there are words which are not desirable as keywords for a domain.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Note that it is in precision that our over\u00c2\u00ad all performance would appear to be poorer than the reported performance of Chang et al., yet based on their published examples, our system appears to be doing better precisionwise.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"As noted in Section 1, our code finds the longest matching entry wnf for query p(wn|s(wn\u22121 f ) The probability p(wn|wn\u22121 f ) is stored with wnf and the backoffs are immediately accessible in the provided state s(wn\u22121 When our code walks the data structure to find wnf , it visits wnn, wnn\u22121, ... , wnf .","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For the experiments, we use a simple preprocessing step.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Thus at each iteration the method induces at most n x k rules, where k is the number of possible labels (k = 3 in the experiments in this paper). step 3.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We will report the evaluation results in the next subsection.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The 2(Yarowsky 95) describes the use of more sophisticated smoothing methods.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"IL+-1Proof: Assume a pair of crossing constituents appears in the output of the constituent voting technique using k parsers.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Each out-of-domain phrase pair is characterized by a set of simple features intended to reflect how useful it will be.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We implement two data structures: PROBING, designed for speed, and TRIE, optimized for memory.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The algorithm in Fig.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The form mnh itself can be read as at least three different verbs (\u201ccounted\u201d, \u201cappointed\u201d, \u201cwas appointed\u201d), a noun (\u201ca portion\u201d), and a possessed noun (\u201cher kind\u201d).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Sie.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Lazy mapping reduces memory requirements by loading pages from disk only as necessary.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We identified three ways that contextual roles can be exploited: (1) by identifying caseframes that co-occur in resolutions, (2) by identifying nouns that co-occur with case- frames and using them to crosscheck anaphor\/candidate compatibility, (3) by identifying semantic classes that co- occur with caseframes and using them to crosscheck anaphor\/candidate compatability.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Schapire and Singer show that the training error is bounded above by Thus, in order to greedily minimize an upper bound on training error, on each iteration we should search for the weak hypothesis ht and the weight at that minimize Z.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In Section 5, we try to compare results of MENE, IdentiFinder, and MENERGI.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"2 70.7 52.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The entries in such a lexicon may be thought of as meaningful surface segments paired up with their PoS tags li = (si, pi), but note that a surface segment s need not be a space-delimited token.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Here we propose a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","label":1},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"1 61.2 43.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"If all three sentences are in the same document, then even a human will find it difficult to classify McCann in (1) into either person or organization, unless there is some other information provided.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"BABAR uses two methods to identify anaphors that can be easily and reliably resolved with their antecedent: lexical seeding and syntactic seeding.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The performance was 80.99% recall and 61.83% precision.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"By contrast, BerkeleyLM\u2019s hash and compressed variants will return incorrect results based on an n \u22121-gram.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"f;g denotes the empty set, where no source sentence position is covered.","label":0},{"source_text":"There is no global pruning.","target_text":"Figure 2: Order in which source positions are visited for the example given in Fig.1.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The second modification is more important, and is discussed in the next section.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"For instance, the sentence Similar improvements in haemoglobin levels were reported in the scientific literature for other epoetins would likely be considered domain-specific despite the presence of general phrases like were reported in.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"M(wi) = Li).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Past work however, has typically associ n = n P (Ti)P (Wi|Ti) = i=1 1 n K n ated these features with token occurrences, typically in an HMM.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"One obvious application is information extraction.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"One conclusion drawn from this annotation effort was that for humans and machines alike, 2 www.sfs.nphil.unituebingen.de\/Elwis\/stts\/ stts.html 3 www.coli.unisb.de\/sfb378\/negra-corpus\/annotate.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We used section 23 as the development set for our combining techniques, and section 22 only for final testing.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, if {N P1, N P2, N P3} are all coreferent, then each NP must be linked to one of the other two NPs.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"What ought to be developed now is an annotation tool that can make use of the format, allow for underspecified annotations and visualize them accordingly.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"(2010) consistently outperforms ours on English, we obtain substantial gains across other languages.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"0 57.2 43.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The extent to which this constraint is enforced varies greatly across existing methods.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"18 77.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For locations, tokens are matched against unigrams, and sequences of two consecutive tokens are matched against bigrams.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"For alif with hamza, normalization can be seen as another level of devocalization.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The intuition here is that the role of a discourse marker can usually be de 9 Both the corpus split and pre-processing code are avail-.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The segmentation chosen is the best path through the WFST, shown in (d).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The model described here thus demonstrates great potential for use in widespread applications.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"That is, we can use the discourse parser on PCC texts, emulating for instance a \u00e2\u0080\u009cco-reference oracle\u00e2\u0080\u009d that adds the information from our co-reference annotations.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Mikheev et al.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"And time is short.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(Other classes handled by the current system are discussed in Section 5.)","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"BABAR merely identifies caseframes that frequently co-occur in coreference resolutions.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"An example of a fairly low-level relation is the affix relation, which holds between a stem morpheme and an affix morpheme, such as f1 -menD (PL).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"58 95.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The effect of the pruning threshold t0 is shown in Table 5.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"For example, CFG's cannot produce trees of the form shown in Figure 1 in which there are nested dependencies between S and NP nodes appearing on the spine of the tree.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The 8 similarity-to-IN features are based on word frequencies and scores from various models trained on the IN corpus: To avoid numerical problems, each feature was normalized by subtracting its mean and dividing by its standard deviation.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"10.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"For Experiment 1 it is meaningless as a baseline, since it would result in 0% accuracy. mation on path labels but drop the information about the syntactic head of the lifted arc, using the label d\u2191 instead of d\u2191h (AuxP\u2191 instead of AuxP\u2191Sb).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The correct resolution in sentence (c) depends on knowledge that kidnappers frequently blindfold their victims.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A secondary reference resolution classifier has information on the class assigned by the primary classifier.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"However, supervised methods rely on labeled training data, which is time-consuming and expensive to generate.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Figure 4 shows a constituent headed by a process nominal with an embedded adjective phrase.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR uses information extraction patterns to identify contextual roles and creates four contextual role knowledge sources using unsupervised learning.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This combination generalizes (2) and (3): we use either at = a to obtain a fixed-weight linear combination, or at = cI(t)\/(cI(t) + 0) to obtain a MAP combination.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Generalizing state minimization, the model could also provide explicit bounds on probability for both backward and forward extension.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"In an experiment on automatic rhetorical parsing, the RST-annotations and PoS tags were used by (Reitter 2003) as a training corpus for statistical classification with Support Vector Machines.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"\"c' 0 + 0 \"0 ' \u00e2\u0080\u00a2 + a n t i g r e e d y x g r e e d y < > c u r r e n t m e t h o d o d i e t . o n l y \u00e2\u0080\u00a2 Taiwan 0 \u00c2\u00b7;; 0 c CD E i5 0\"' 9 9 \u00e2\u0080\u00a2 Mainland \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 -0.30.20.1 0.0 0.1 0.2 Dimension 1 (62%) Figure 7 Classical metric multidimensional scaling of distance matrix, showing the two most significant dimensions.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"This PP modifies another NP, whose head is a singular noun.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"This supports our main thesis that decisions taken by single, improved, grammar are beneficial for both tasks.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Equ.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We extend Subramanya et al.\u2019s intuitions to our bilingual setup.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"0 D\/ nc 5.0 The minimal dictionary encoding this information is represented by the WFST in Figure 2(a).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Almost all annotators expressed their preference to move to a ranking-based evaluation in the future.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The simplest approach involves scoring the various analyses by costs based on word frequency, and picking the lowest cost path; variants of this approach have been described in Chang, Chen, and Chen (1991) and Chang and Chen (1993).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Given that weights on all outgoing arcs sum up to one, weights induce a probability distribution on the lattice paths.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"An example of a fairly low-level relation is the affix relation, which holds between a stem morpheme and an affix morpheme, such as f1 -menD (PL).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Modifying the Berkeley parser for Arabic is straightforward.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"It falls short of the \u201cProjection\u201d baseline for German, but is statistically indistinguishable in terms of accuracy.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For parsing, this is a mistake, especially in the case of interrogatives.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The semantic caseframe expectations are used in two ways.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"However, for multinomial models like our LMs and TMs, there is a one to one correspondence between instances and features, eg the correspondence between a phrase pair (s, t) and its conditional multinomial probability p(s1t).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Discriminative Instance Weighting for Domain Adaptation in Statistical Machine Translation","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"3.3 Evaluation Results.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"These systems rely on a training corpus that has been manually annotated with coreference links.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Dictionaries: Due to the limited amount of training material, name dictionaries have been found to be useful in the named entity task.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"7 Conclusion and Future Work.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Unlike the WSJ corpus which has a high frequency of rules like VP \u00e2\u0086\u0092VB PP, Arabic verb phrases usually have lexi calized intervening nodes (e.g., NP subjects and direct objects).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Vocabulary lookup is a hash table mapping from word to vocabulary index.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"These are shown, with their associated costs, as follows: ABj nc 4.0 AB C\/jj 6.0 CD \/vb 5.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Call the crossing constituents A and B.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We report results for the best and median hyperparameter settings obtained in this way.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"With each iteration more examples are assigned labels by both classifiers, while a high level of agreement (> 94%) is maintained between them.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The treebank has two versions, v1.0 and v2.0, containing 5001 and 6501 sentences respectively.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While building a machine translation system is a serious undertaking, in future we hope to attract more newcomers to the field by keeping the barrier of entry as low as possible.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In addition to a heuristic based on decision list learning, we also presented a boosting-like framework that builds on ideas from (Blum and Mitchell 98).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"For simplicity, we assume that OUT is homogeneous.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In this paper we present a stochastic finite-state model for segmenting Chinese text into words, both words found in a (static) lexicon as well as words derived via the above-mentioned productive processes.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"A moment's reflection will reveal that things are not quite that simple.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"This work was supported in part by the National Science Foundation under grant IRI9704240.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"To approximate these baselines, we implemented a very simple sentence selection algorithm in which parallel sentence pairs from OUT are ranked by the perplexity of their target half according to the IN language model.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Richer tag sets have been suggested for modeling morphologically complex distinctions (Diab, 2007), but we find that linguistically rich tag sets do not help parsing.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Hence, for each token, one of the four features zone-TXT, zone- HL, zone-DATELINE, or zone-DD is set to 1, and the other 3 are set to 0.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We pick the parse that is most similar to the other parses by choosing the one with the highest sum of pairwise similarities.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The second model (+PRIOR) utilizes the independent prior over type-level tag assignments P (T |\u00cf\u0088).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"This intuition is born out by the experimental results.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We further thank Khalil Simaan (ILLCUvA) for his careful advise concerning the formal details of the proposal.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This is to allow for fair comparison between the statistical method and GR, which is also purely dictionary-based.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"37 79.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Figure 3 Figure 1.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The larger sets are more accurate than the small sets.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"(2009), who also incorporate a sparsity constraint, but does via altering the model objective using posterior regularization.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The size of TRIE is particularly sensitive to F1092 c11, so vocabulary filtering is quite effective at reducing model size.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"A generative model was applied (similar to naive Bayes) with the three labels as hidden vanables on unlabeled examples, and observed variables on (seed) labeled examples.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J) and the lexicon probability p(fbi jei) are obtained by relative frequency estimates from the Viterbi alignment path after the final training iteration.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Finally, we note that Jiang\u2019s instance-weighting framework is broader than we have presented above, encompassing among other possibilities the use of unlabelled IN data, which is applicable to SMT settings where source-only IN corpora are available.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Church and Hanks [1989]), and we have used lists of character pairs ranked by mutual information to expand our own dictionary.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For the purposes of EM, the &quot;observed&quot; data is {(xi, Ya\u2022 \u2022 \u2022 (xrn, Yrn), xfil, and the hidden data is {ym+i y}.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The computing time, the number of search errors, and the multi-reference WER (mWER) are shown as a function of t0.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"This locally normalized log-linear model can look at various aspects of the observation x, incorporating overlapping features of the observation.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"As can be seen in Table 4, our training data is a lot less than those used by MENE and IdentiFinder3.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For MUC7, there are also no published results on systems trained on only the official training data of 200 aviation disaster articles.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For example, if is found in the list of person first names, the feature PersonFirstName is set to 1.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"wo rd => na m e 2.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Personal names such as 00, 3R; zhoulenl-lai2 'Zhou Enlai.'","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"can expect famous names like Zhou Enlai's to be in many dictionaries, but names such as :fi lf;f; shi2jil-lin2, the name of the second author of this paper, will not be found in any dictionary.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Feature selection is implemented using a feature cutoff: features seen less than a small count during training will not be used.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Analysis of the data revealed that the contextual role knowledge is especially helpful for resolving pronouns because, in general, they are semantically weaker than definite NPs.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The unlabeled data gives many such &quot;hints&quot; that two features should predict the same label, and these hints turn out to be surprisingly useful when building a classifier.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"With this restriction the resulting tree sets will have independent paths.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"2.1.1 Lexical Seeding It is generally not safe to assume that multiple occurrences of a noun phrase refer to the same entity.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Segments with the same surface form but different PoS tags are treated as different lexemes, and are represented as separate arcs (e.g. the two arcs labeled neim from node 6 to 7).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Note that the sets of possible classifiers for a given noun can easily be encoded on that noun by grammatical features, which can be referred to by finite-state grammatical rules.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The negative logarithm of t0 is reported.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The dictionary sizes reported in the literature range from 17,000 to 125,000 entries, and it seems reasonable to assume that the coverage of the base dictionary constitutes a major factor in the performance of the various approaches, possibly more important than the particular set of methods used in the segmentation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Chang of Tsinghua University, Taiwan, R.O.C., for kindly providing us with the name corpora.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The approach builds from an initial seed set for a category, and is quite similar to the decision list approach described in (Yarowsky 95).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A totally non\u00c2\u00ad stochastic rule-based system such as Wang, Li, and Chang's will generally succeed in such cases, but of course runs the risk of overgeneration wherever the single-hanzi word is really intended.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If it is made up of all capital letters, then (allCaps, zone) is set to 1.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"ogy (Koskenniemi 1983; Antworth 1990; Tzoukermann and Liberman 1990; Karttunen, Kaplan, and Zaenen 1992; Sproat 1992); we represent the fact that ir, attaches to nouns by allowing t:-transitions from the final states of all noun entries, to the initial state of the sub-WFST representing f,.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The structure uses linear probing hash tables and is designed for speed.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"In the cases where isolated constituent precision is larger than 0.5 the affected portion of the hypotheses is negligible.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The corpus-based statistical parsing community has many fast and accurate automated parsing systems, including systems produced by Collins (1997), Charniak (1997) and Ratnaparkhi (1997).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"(2009) study related but different multilingual grammar and tagger induction tasks, where it is assumed that no labeled data at all is available.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The corpus has been annotated with six different types of information, which are characterized in the following subsections.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Employing a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions is not only theoretically clean and linguistically justified and but also probabilistically apropriate and empirically sound.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Recent research has demonstrated that incorporating this sparsity constraint improves tagging accuracy.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":".","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Projectivizing a dependency graph by lifting nonprojective arcs is a nondeterministic operation in the general case.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"One may argue with these efforts on normalization, and ultimately their value should be assessed by assessing their impact on inter-annotator agreement.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The state-to-tag mapping is obtained from the best hyperparameter setting for 11 mapping shown in Table 3.","label":0},{"source_text":"There is no global pruning.","target_text":"2) An improved language model, which takes into account syntactic structure, e.g. to ensure that a proper English verbgroup is generated.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Within the RST \u00e2\u0080\u009cuser community\u00e2\u0080\u009d there has also been discussion whether two levels of discourse structure should not be systematically distinguished (intentional versus informational).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We performed three experiments to evaluate our techniques.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Table 2 shows single-threaded results, mostly for comparison to IRSTLM, and Table 3 shows multi-threaded results.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Although this feature helps, we encounter one consequence of variable word order.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"suffixes (e.g., \u00ef\u00bf\u00bd=) Other notable parameters are second order vertical Markovization and marking of unary rules.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"2 for the accuracy of the different methods.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The present proposal falls into the last group.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This is because our corpus is not annotated, and hence does not distinguish between the various words represented by homographs, such as, which could be \/adv jiangl 'be about to' orInc jiang4 '(military) general'-as in 1j\\xiao3jiang4 'little general.'","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"To differentiate between the coordinating and discourse separator functions of conjunctions (Table 3), we mark each CC with the label of its right sister (splitCC).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"BerkeleyLM uses states to optimistically search for longer n-gram matches first and must perform twice as many random accesses to retrieve backoff information.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We believe it is natural for authors to use abbreviations in subsequent mentions of a named entity (i.e., first \u00e2\u0080\u009cPresident George Bush\u00e2\u0080\u009d then \u00e2\u0080\u009cBush\u00e2\u0080\u009d).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"4 7 . 3 8 . 9 2 8 . 8 2 0 . 7 3 2 . 3 3 5 . 2 2 9 . 6 2 7 . 6 1 4 . 2 4 2 . 8 4 5 . 9 4 4 . 3 6 0 . 6 6 1 . 5 4 9 . 9 3 3 . 9 Table 6: Type-level Results: Each cell report the type- level accuracy computed against the most frequent tag of each word type.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"This analyzer setting is similar to that of (Cohen and Smith, 2007), and models using it are denoted nohsp, Parser and Grammar We used BitPar (Schmid, 2004), an efficient general purpose parser,10 together with various treebank grammars to parse the input sentences and propose compatible morphological segmentation and syntactic analysis.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We can use the preceding linguistic and annotation insights to build a manually annotated Arabic grammar in the manner of Klein and Manning (2003).","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In the following, we use the notation wi wj to mean that (wi, r, wj) E A; r we also use wi wj to denote an arc with unspecified label and wi \u2014*\u2217 wj for the reflexive and transitive closure of the (unlabeled) arc relation.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We have already mentioned the closely related work by Matsoukas et al (2009) on discriminative corpus weighting, and Jiang and Zhai (2007) on (nondiscriminative) instance weighting.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Segmental morphology Hebrew consists of seven particles m(\u201cfrom\u201d) f(\u201cwhen\u201d\/\u201cwho\u201d\/\u201cthat\u201d) h(\u201cthe\u201d) w(\u201cand\u201d) k(\u201clike\u201d) l(\u201cto\u201d) and b(\u201cin\u201d). which may never appear in isolation and must always attach as prefixes to the following open-class category item we refer to as stem.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The tree denoting this derivation of 7 is rooted with a node labeled 7' having k subtrees for the derivations of 71, ... ,7a.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"There is a fairly large body of work on SMT adaptation.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The extent to which this constraint is enforced varies greatly across existing methods.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The following three sections elaborate these different stages is more detail.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Out-of-Vocabulary: We derived a lexicon list from WordNet 1.6, and words that are not found in this list have a feature out-of-vocabulary set to 1.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This range is collapsed to a number of buckets, typically by taking the hash modulo the number of buckets.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The 1st block contains the simple baselines from section 2.1.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"This section describes AdaBoost, which is the basis for the CoBoost algorithm.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"To explicitly handle the word reordering between words in source and target language, we use the concept of the so-called inverted alignments as given in (Ney et al., 2000).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"It's not clear how to apply these methods in the unsupervised case, as they required cross-validation techniques: for this reason we use the simpler smoothing method shown here. input to the unsupervised algorithm is an initial, &quot;seed&quot; set of rules.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"By aggregating the POS labels of the English tokens to types, we can generate label distributions for the English vertices.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For example, take ..., says Maury Cooper, a vice president at S.&P.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Specifically, for the ith word type, the set of token-level tags associated with token occurrences of this word, denoted t(i), must all take the value Ti to have nonzero mass. Thus in the context of Gibbs sampling, if we want to block sample Ti with t(i), we only need sample values for Ti and consider this setting of t(i).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"For example, we might have VP \u00e2\u0086\u0092 VB NP PP, where the NP is the subject.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The following algorithm was then used to induce new rules: Let Count' (x) be the number of times feature x is seen with some known label in the training data.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The Danish Dependency Treebank (DDT) comprises about 100K words of text selected from the Danish PAROLE corpus, with annotation of primary and secondary dependencies (Kromann, 2003).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The theory has also been validated empirically.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The range of the score is between 0 and 1 (higher is better).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"2.1 Overview.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We describe a novel approach for inducing unsupervised part-of-speech taggers for languages that have no labeled training data, but have translated text in a resource-rich language.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"For one thing, it is not clear who is to receive settlements or what should happen in case not enough teachers accept the offer of early retirement.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Recent results (e.g., (Yarowsky 95; Brill 95; Blum and Mitchell 98)) have suggested that unlabeled data can be used quite profitably in reducing the need for supervision.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Upon identifying an anaphoric expression (currently restricted to: pronouns, prepositional adverbs, definite noun phrases), the an- notator first marks the antecedent expression (currently restricted to: various kinds of noun phrases, prepositional phrases, verb phrases, sentences) and then establishes the link between the two.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Decreasing the threshold results in higher mWER due to additional search errors.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Evalb, the standard parsing metric, is biased toward such corpora (Sampson and Babarczy, 2003).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"MENE has only been tested on MUC7.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Human evaluation is one way to distinguish between the two cases.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"It was also proposed to allow annotators to skip sentences that they are unable to judge.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We outlined the definition of a family of constrained grammatical formalisms, called Linear Context-Free Rewriting Systems.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In principle, it would be possible to encode the exact position of the syntactic head in the label of the arc from the linear head, but this would give a potentially infinite set of arc labels and would make the training of the parser very hard.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Evaluation within a set The evaluation of paraphrases within a set of phrases which share a keyword is illustrated in Figure 4.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In general, several gross corpus statistics favor the ATB, so other factors must contribute to parsing underperformance.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The Wang, Li, and Chang system fails on fragment (b) because their system lacks the word youlyoul 'soberly' and misinterpreted the thus isolated first youl as being the final hanzi of the preceding name; similarly our system failed in fragment (h) since it is missing the abbreviation i:lJI!","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"For parsing, the most challenging form of ambiguity occurs at the discourse level.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Since our objective is to compare distributions of bracketing discrepancies, we do not use heuristics to prune the set of nuclei.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Except for the left3The graphs satisfy all the well-formedness conditions given in section 2 except (possibly) connectedness.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"immediately by a Romanization into the pinyin transliteration scheme; numerals following each pinyin syllable represent tones.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Blum and Mitchell 98) offer a promising formulation of redundancy, also prove some results about how the use of unlabeled examples can help classification, and suggest an objective function when training with unlabeled examples.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"It was developed in response to the non-terminal\/terminal bias of Evalb, but Clegg and Shepherd (2005) showed that it is also a valuable diagnostic tool for trees with complex deep structures such as those found in the ATB.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"However, we do not explore this possibility in the current work.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our work is motivated by the observation that contextual roles can be critically important in determining the referent of a noun phrase.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Non-empty buckets contain an entry belonging to them or to a preceding bucket where a conflict occurred.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"In the initial release of the ATB, inter-annotator agreement was inferior to other LDC treebanks (Maamouri et al., 2008).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"I \u00e2\u0080\u00a2 JAPANS :rl4 .\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\)\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"o\u00c2\u00b7'\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\:J\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7 \u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7'\\; . '.:: ..........0 6.51 9.51 : jj \/ JAPANESE OCTOPUS 10\u00c2\u00b728i\u00c2\u00a3 :_nc HOW SAY f B :rl4 :il: :wen2 t '- \u00e2\u0080\u00a2 :zhang!","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Table 3: Dev set frequencies for the two most significant discourse markers in Arabic are skewed toward analysis as a conjunction.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"There is no global pruning.","target_text":"Future extensions of the system might include: 1) An extended translation model, where we use more context to predict a source word.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Figure 2: An ATB sample from the human evaluation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"To evaluate proper-name identification, we randomly se\u00c2\u00ad lected 186 sentences containing 12,000 hanzi from our test corpus and segmented the text automatically, tagging personal names; note that for names, there is always a sin\u00c2\u00ad gle unambiguous answer, unlike the more general question of which segmentation is correct.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Note that in our model, conditioned on T , there is precisely one t which has nonzero probability for the token component, since for each word, exactly one \u00ce\u00b8t has support.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The most frequently occurring last words of organization names in cslist are compiled into a list of corporate suffixes, Corporate-Suffix- List.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Other errors include NE tagging errors and errors due to a phrase which includes other NEs.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"For each language and setting, we report one-to-one (11) and many- to-one (m-1) accuracies.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Many researchers have developed coreference resolvers, so we will only discuss the methods that are most closely related to BABAR.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This is in general very difficult, given the extremely free manner in which Chinese given names are formed, and given that in these cases we lack even a family name to give the model confidence that it is identifying a name.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Unlabeled examples in the named-entity classification problem can reduce the need for supervision to a handful of seed rules.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"needs to be in initCaps to be considered for this feature.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The results clearly indicate increased variation in the ATB relative to the WSJ, but care should be taken in assessing the magnitude of the difference.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Such analyzers propose multiple segmentation possibilities and their corresponding analyses for a token in isolation but have no means to determine the most likely ones.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Yet, some hanzi are far more probable in women's names than they are in men's names, and there is a similar list of male-oriented hanzi: mixing hanzi from these two lists is generally less likely than would be predicted by the independence model.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Hence, s < k. But by addition of the votes on the two parses, s > 2N-11> k, a contradiction.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"971,746 sentences of New York Times text were parsed using the parser of (Collins 96).1 Word sequences that met the following criteria were then extracted as named entity examples: whose head is a singular noun (tagged NN).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"These results are promising and there are several avenues for improving on these results.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Other strategies could readily 6 As a reviewer has pointed out, it should be made clear that the function for computing the best path is. an instance of the Viterbi algorithm.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"4 70.4 46.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"For each pair we also record the context, i.e. the phrase between the two NEs (Step1).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Table 2 shows single-threaded results, mostly for comparison to IRSTLM, and Table 3 shows multi-threaded results.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"For the remaining arcs, if the segment is in fact a known lexeme it is tagged as usual, but for the OOV arcs which are valid Hebrew entries lacking tags assignment, we assign all possible tags and then simulate a grammatical constraint.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The code is opensource, has minimal dependencies, and offers both C++ and Java interfaces for integration.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"We also see that the increase in the size of the label sets for Head and Head+Path is far below the theoretical upper bounds given in Table 1.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The scoping heuristics are based on the anaphor type: for reflexive pronouns the scope is the current clause, for relative pronouns it is the prior clause following its VP, for personal pronouns it is the anaphor\u00e2\u0080\u0099s sentence and two preceding sentences, and for definite NPs it is the anaphor\u00e2\u0080\u0099s sentence and eight preceding sentences.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"After the first step towards breadth had been taken with the PoS-tagging, RST annotation, and URML conversion of the entire corpus of 170 texts12 , emphasis shifted towards depth.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We are currently exploring other methods that employ similar ideas and their formal properties.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"1 61.7 37.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In IE, creating the patterns which express the requested scenario, e.g. \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d or \u00e2\u0080\u009ccorporate merger and acquisition\u00e2\u0080\u009d is regarded as the hardest task.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"52 15.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(5) and ht into Equ.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For each 13 Of course, this weighting makes the PCFG an improper distribution.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The iw all map variant uses a non-0 y weight on a uniform prior in p,,(s t), and outperforms a version with y = 0 (iw all) and the \u201cflattened\u201d variant described in section 3.2.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"This approach needs a phrase as an initial seed and thus the possible relationships to be extracted are naturally limited.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"As noted, this sentence consists of four words, namely B X ri4wen2 'Japanese,' :\u00c2\u00a5, zhanglyu2 'octopus\/ :&P:l zen3me0 'how,' and IDt shuol 'say.'","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Except for the left3The graphs satisfy all the well-formedness conditions given in section 2 except (possibly) connectedness.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Following Sproat and Shih (1990), performance for Chinese segmentation systems is generally reported in terms of the dual measures of precision and recalP It is fairly standard to report precision and recall scores in the mid to high 90% range.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The problem is to store these two values for a large and sparse set of n-grams in a way that makes queries efficient.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"we perform five runs with different random initialization of sampling state.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The first value reports resident size after loading; the second is the gap between post-loading resident memory and peak virtual memory.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The features are used to represent each example for the learning algorithm.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"One implementation issue deserves some elaboration.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"In the cases where isolated constituent precision is larger than 0.5 the affected portion of the hypotheses is negligible.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"2.1 Overview.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We will refer to our system as MENERGI (Maximum Entropy Named Entity Recognizer using Global Information).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The binary language model from Section 5.2 and text phrase table were forced into disk cache before each run.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The ability to redistribute belief values across sets rather than individual hypotheses is key.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A configuration of M consists of a state of the finite control, the nonblank contents of the input tape and k work tapes, and the position of each head.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"constitute names, since we have only their segmentation, not the actual classification of the segmented words.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"4 To be sure, it is not always true that a hanzi represents a syllable or that it represents a morpheme.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The learned information was recycled back into the resolver to improve its performance.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"So, there is a limitation that IE can only be performed for a predefined task, like \u00e2\u0080\u009ccorporate mergers\u00e2\u0080\u009d or \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Pseudo-labels are formed by taking seed labels on the labeled examples, and the output of the fixed classifier on the unlabeled examples.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"First, we aim to explicitly characterize examples from OUT as belonging to general language or not.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The \u201cNo LP\u201d model does not outperform direct projection for German and Greek, but performs better for six out of eight languages.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The computing time, the number of search errors, and the multi-reference WER (mWER) are shown as a function of t0.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Yarowsky 95) describes an algorithm for word-sense disambiguation that exploits redundancy in contextual features, and gives impressive performance.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Nonstochastic lexical-knowledge-based approaches have been much more numer\u00c2\u00ad ous.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For example, kidnapping victims should be extracted from the subject of the verb \u00e2\u0080\u009ckidnapped\u00e2\u0080\u009d when it occurs in the passive voice (the shorthand representation of this pattern would be \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"For each cell, the first row corresponds to the result using the best hyperparameter choice, where best is defined by the 11 metric.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"58 95.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"These enable much larger models in memory, compensating for lost accuracy.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In the remainder of the paper, we outline how a class of Linear Context-Free Rewriting Systems (LCFRS's) may be defined and sketch how semilinearity and polynomial recognition of these systems follows.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"When OUT is large and distinct, its contribution can be controlled by training separate IN and OUT models, and weighting their combination.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Thus we have some confidence that our own performance is at least as good as that of Chang et al.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Unfortunately, the best completely unsupervised English POS tagger (that does not make use of a tagging dictionary) reaches only 76.1% accuracy (Christodoulopoulos et al., 2010), making its practical usability questionable at best.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Other errors include NE tagging errors and errors due to a phrase which includes other NEs.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The second model (+PRIOR) utilizes the independent prior over type-level tag assignments P (T |\u00cf\u0088).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Our TRIE implements the popular reverse trie, in which the last word of an n-gram is looked up first, as do SRILM, IRSTLM\u2019s inverted variant, and BerkeleyLM except for the scrolling variant.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This departure from the traditional token-based tagging approach allows us to explicitly capture type- level distributional properties of valid POS tag as signments as part of the model.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The verbal reading arises when the maSdar has an NP argument which, in vocalized text, is marked in the accusative case.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The alignment model uses two kinds of parameters: alignment probabilities p(aj jaj\u00f4\u0080\u0080\u00801; I; J), where the probability of alignment aj for position j depends on the previous alignment position aj\u00f4\u0080\u0080\u00801 (Ney et al., 2000) and lexicon probabilities p(fj jeaj ).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, if CFLex determines that the log- likelihood statistic for the co-occurrence of a particular noun and caseframe corresponds to the 90% confidence level, then CFLex returns .90 as its belief that the anaphor and candidate are coreferent.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"computing the precision of the other's judgments relative to this standard.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We then extract a set of possible tags tx(y) by eliminating labels whose probability is below a threshold value \u03c4: We describe how we choose \u03c4 in \u00a76.4.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Part of the gap between resident and virtual memory is due to the time at which data was collected.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"3.4 Salience-based text generation.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The second main result is that the pseudo-projective approach to parsing (using special arc labels to guide an inverse transformation) gives a further improvement of about one percentage point on attachment score.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"First, we will describe their method and compare it with our method.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Assume that the two classifiers are &quot;rote learners&quot;: that is, 1.1 and 12 are defined through look-up tables that list a label for each member of X1 or X2.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"This measure has the advantage of being completely automatic.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Thus, provided at least this amount of IN data is available\u2014as it is in our setting\u2014adapting these weights is straightforward.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"There are thus some very good reasons why segmentation into words is an important task.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"This is a unique object for which we are able to define a proper probability model.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"With the additional assumptions, inspired by Rounds (1985), we can show that members of this class can be recognized in polynomial time.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"1 61.7 37.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The new algorithm, which we call CoBoost, uses labeled and unlabeled data and builds two classifiers in parallel.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We also mark all tags that dominate a word with the feminine ending :: taa mar buuTa (markFeminine).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"our full model yields 39.3% average error reduction across languages when compared to the basic configuration (1TW).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Apart from MERT difficulties, a conceptual problem with log-linear combination is that it multiplies feature probabilities, essentially forcing different features to agree on high-scoring candidates.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The increase is generally higher for PDT than for DDT, which indicates a greater diversity in non-projective constructions.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Next, for each pair of NE categories, we collect all the contexts and find the keywords which are topical for that NE category pair.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"40 75.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"92 77.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"However, TRIE partitions storage by n-gram length, so walking the trie reads N disjoint pages.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We conjecture that this trend may continue by incorporating additional information, e.g., three-dimensional models as proposed by Tsarfaty and Sima\u2019an (2007).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This highly effective approach is not directly applicable to the multinomial models used for core SMT components, which have no natural method for combining split features, so we rely on an instance-weighting approach (Jiang and Zhai, 2007) to downweight domain-specific examples in OUT.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Affix Pron Base category N found N missed (recall) N correct (precision) t,-,7 The second issue is that rare family names can be responsible for overgeneration, especially if these names are otherwise common as single-hanzi words.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This technique was introduced by Clarkson and Rosenfeld (1997) and is also implemented by IRSTLM and BerkeleyLM\u2019s compressed option.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Thus, if one wants to segment words-for any purpose-from Chinese sentences, one faces a more difficult task than one does in English since one cannot use spacing as a guide.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"This is in contrast to dependency treebanks, e.g.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"from the subset of the United Informatics corpus not used in the training of the models.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Our annotators pointed out that very often they made almost random decisions as to what relation to choose, and where to locate the boundary of a span.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Some of these approaches (e.g., Lin, Chiang, and Su [1993]) attempt to identify unknown words, but do not ac\u00c2\u00ad tually tag the words as belonging to one or another class of expression.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"E.g. when 'Zahnarzttermin' is aligned to dentist's, the extended lexicon model might learn that 'Zahnarzttermin' actuallyhas to be aligned to both dentist's and ap pointment.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"This offers the well-known advantages for inter- changability, but it raises the question of how to query the corpus across levels of annotation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"One hopes that such a corpus will be forth\u00c2\u00ad coming.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Using these states, we define partial hypothesis extensions, which are of the following type: (S0;C n fjg; j0) !","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Restrictions: Quasi-monotone Search The above search space is still too large to allow the translation of a medium length input sentence.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We will refer to our system as MENERGI (Maximum Entropy Named Entity Recognizer using Global Information).","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"information structure.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Bikel 75 training trees 5000 10000 15000 Figure 3: Dev set learning curves for sentence lengths \u00e2\u0089\u00a4 70.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"72 78.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Finally, we note that simple weighting gives nearly a 2% F1 improvement, whereas Goldberg and Tsarfaty (2008) found that unweighted lattices were more effective for Hebrew.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In this paper, we have presented a new, e\u00c3\u0086cient DP-based search procedure for statistical machine translation.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"1 61.2 43.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Links can be of two different kinds: anaphoric or bridging (definite noun phrases picking up an antecedent via world-knowledge).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The most frequent NE category pairs are \u00e2\u0080\u009cPerson - Person (209,236), followed by \u00e2\u0080\u009cCountry - Coun- try\u00e2\u0080\u009d (95,123) and \u00e2\u0080\u009cPerson - Country\u00e2\u0080\u009d (75,509).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"In our case multi-threading is trivial because our data structures are read-only and uncached.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Also, we don\u00e2\u0080\u0099t know how many such paraphrase sets are necessary to cover even some everyday things or events.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For instance, for TTS it is necessary to know that a particular sequence of hanzi is of a particular category because that knowl\u00c2\u00ad edge could affect the pronunciation; consider, for example the issues surrounding the pronunciation of ganl I qian2 discussed in Section 1.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Two lists, Corporate-Suffix-List (for corporate suffixes) and Person-Prefix-List (for person prefixes), are collected from the training data.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This flexibility, along with the simplicity of implementation and expansion, makes this framework an attractive base for continued research.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This is manifest in the lexical choices but 1 www.coli.unisb.de\/\u00e2\u0088\u00bcthorsten\/tnt\/ Dagmar Ziegler is up to her neck in debt.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Our system fails in (a) because of$ shenl, a rare family name; the system identifies it as a family name, whereas it should be analyzed as part of the given name.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"These knowledge sources determine whether the contexts surrounding an anaphor and antecedent are compatible.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"However, it is unlikely that other occurrences of News Broadcasting Corp. in the same document also co-occur with Even.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"These make left-to-right query patterns convenient, as the application need only provide a state and the word to append, then use the returned state to append another word, etc.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The type-level tag assignments T generate features associated with word types W . The tag assignments constrain the HMM emission parameters \u00ce\u00b8.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"As such, global information from the whole context of a document is important to more accurately recognize named entities.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Although these existential NPs do not need a prior referent, they may occur multiple times in a document.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We have presented a method for unsupervised part- of-speech tagging that considers a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The Bayes models were able to achieve significantly higher precision than their non-parametric counterparts.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The details of the transformation procedure are slightly different depending on the encoding schemes: d\u2191h let the linear head be the syntactic head). target arc must have the form wl \u2212\u2192 wm; if no target arc is found, Head is used as backoff. must have the form wl \u2212\u2192 wm and no outgoing arcs of the form wm p'\u2193 \u2212\u2192 wo; no backoff.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In this work we extended the AdaBoost.MH (Schapire and Singer 98) algorithm to the cotraining case.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Instead, we resort to an iterative update based method.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Subsets C of increasing cardinality c are processed.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"For developers of Statistical Machine Translation (SMT) systems, an additional complication is the heterogeneous nature of SMT components (word-alignment model, language model, translation model, etc.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"As a result, its POS tag needs to be induced in the \u201cNo LP\u201d case, while the 11A word level paired-t-test is significant at p < 0.01 for Danish, Greek, Italian, Portuguese, Spanish and Swedish, and p < 0.05 for Dutch. correct tag is available as a constraint feature in the \u201cWith LP\u201d case.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This makes memory usage comparable to our PROBING model.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The PROBING model can perform optimistic searches by jumping to any n-gram without needing state and without any additional memory.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We used it to score all phrase pairs in the OUT table, in order to provide a feature for the instance-weighting model.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For E(ni1s), then, we substitute a smooth S against the number of class elements.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"A receives a votes, and B receives b votes.","label":0},{"source_text":"There is no global pruning.","target_text":"This approach is compared to another reordering scheme presented in (Berger et al., 1996).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"G1 and G2 are hanzi, we can estimate the probability of the sequence being a name as the product of: \u00e2\u0080\u00a2 the probability that a word chosen randomly from a text will be a name-p(rule 1), and \u00e2\u0080\u00a2 the probability that the name is of the form 1hanzi-family 2hanzi-given-p(rule 2), and \u00e2\u0080\u00a2 the probability that the family name is the particular hanzi F1-p(rule 6), and \u00e2\u0080\u00a2 the probability that the given name consists of the particular hanzi G1 and G2-p(rule 9) This model is essentially the one proposed in Chang et al.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"If the log backoff of wnf is also zero (it may not be in filtered models), then wf should be omitted from the state.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Fourth, we show how to build better models for three different parsers.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Similar advances have been made in machine translation (Frederking and Nirenburg, 1994), speech recognition (Fiscus, 1997) and named entity recognition (Borthwick et al., 1998).","label":0},{"source_text":"Here we present two algorithms.","target_text":"Note that in our formalism a weakhypothesis can abstain.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Proper names are assumed to be coreferent if they match exactly, or if they closely match based on a few heuristics.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Adam Pauls provided a pre-release comparison to BerkeleyLM and an initial Java interface.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Compared to decoding, this task is cache-unfriendly in that repeated queries happen only as they naturally occur in text.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"(Webber et al., 2003)).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The final estimating equation is then: (3) Since the total of all these class estimates was about 10% off from the Turing estimate n1\/N for the probability of all unseen hanzi, we renormalized the estimates so that they would sum to n 1jN.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"i..f,..","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Other approaches encode sparsity as a soft constraint.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"There any many techniques for improving language model speed and reducing memory consumption.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We define a symmetric similarity function K(uZ7 uj) over two foreign language vertices uZ7 uj E Vf based on the co-occurrence statistics of the nine feature concepts given in Table 1.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This significantly underperforms log-linear combination.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Formally, let el (62) be the number of classification errors of the first (second) learner on the training data, and let Eco be the number of unlabeled examples on which the two classifiers disagree.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The highestorder N-gram array omits backoff and the index, since these are not applicable.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"2.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"SEGTok(noH) is the segmentation accuracy ignoring mistakes involving the implicit definite article h.11 To evaluate our performance on the tagging task we report CPOS and FPOS corresponding to coarse- and fine-grained PoS tagging results (F1) measure.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Collisions between two keys in the table can be identified at model building time.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Section 2.2 then describes our representation for contextual roles and four types of contextual role knowledge that are learned from the training examples.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"While there are other obstacles to completing this idea, we believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The development of a na\u00efve Bayes classifier involves learning how much each parser should be trusted for the decisions it makes.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"After the first step towards breadth had been taken with the PoS-tagging, RST annotation, and URML conversion of the entire corpus of 170 texts12 , emphasis shifted towards depth.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Recent work has made significant progress on unsupervised POS tagging (Me\u00c2\u00b4rialdo, 1994; Smith and Eisner, 2005; Haghighi and Klein, 2006; Johnson,2007; Goldwater and Griffiths, 2007; Gao and John son, 2008; Ravi and Knight, 2009).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"of Arabic.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The first shows how constituent features and context do not help in deciding which parser to trust.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Besides the lack of a clear definition of what constitutes a correct segmentation for a given Chinese sentence, there is the more general issue that the test corpora used in these evaluations differ from system to system, so meaningful comparison between systems is rendered even more difficult.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"5.2 Setup.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Cluster phrases based on Links We now have a set of phrases which share a keyword.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Hence, we take the probability of the event fmnh analyzed as REL VB to be This means that we generate f and mnh independently depending on their corresponding PoS tags, and the context (as well as the syntactic relation between the two) is modeled via the derivation resulting in a sequence REL VB spanning the form fmnh. based on linear context.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(See Sproat and Shih 1995.)","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Using the concept of inverted alignments, we explicitly take care of the coverage constraint by introducing a coverage set C of source sentence positions that have been already processed.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"input token, the segmentation is then performed deterministically given the 1-best analysis.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"1 2 3.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Cluster phrases based on Links We now have a set of phrases which share a keyword.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"First, we will describe their method and compare it with our method.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For example, if a token starts with a capital letter and ends with a period (such as Mr.), then the feature InitCapPeriod is set to 1, etc. First Word: This feature group contains only one feature firstword.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In the natural disasters domain, agents are often forces of nature, such as hurricanes or wildfires.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Sparse lookup is a key subproblem of language model queries.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We have presented two general approaches to studying parser combination: parser switching and parse hybridization.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Reflexive pronouns with only 1 NP in scope..","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Daum\u00b4e (2007) applies a related idea in a simpler way, by splitting features into general and domain-specific versions.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"First, we parsed the training corpus, collected all the noun phrases, and looked up each head noun in WordNet (Miller, 1990).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"8 We use head-finding rules specified by a native speaker.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, it is possible to personify any noun, so in children's stories or fables, i\u00c2\u00a5JJ1l.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Eight out of the thirteen errors in the high frequency phrases in the CC-domain are the phrases in \u00e2\u0080\u009cagree\u00e2\u0080\u009d.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Syntactic universals are a well studied concept in linguistics (Carnie, 2002; Newmeyer, 2005), and were recently used in similar form by Naseem et al. (2010) for multilingual grammar induction.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"4.2 Global Features.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"It was developed in response to the non-terminal\/terminal bias of Evalb, but Clegg and Shepherd (2005) showed that it is also a valuable diagnostic tool for trees with complex deep structures such as those found in the ATB.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"termined by the category of the word that follows it.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For a given \"word\" in the automatic segmentation, if at least k of the hu\u00c2\u00ad man judges agree that this is a word, then that word is considered to be correct.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"This is the first set that gives us a fair evaluation of the Bayes models, and the Bayes switching model performs significantly better than its non-parametric counterpart.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We do not adapt the alignment procedure for generating the phrase table from which the TM distributions are derived.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This suggests a strategy: run interpolation search until the range narrows to 4096 or fewer entries, then switch to binary search.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"In both cases, SRILM walks its trie an additional time to minimize context as mentioned in Section 4.1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"What is important and is not expressed by the notation is the so-called coverage constraint: each source position j should be 'hit' exactly once by the path of the inverted alignment bI 1 = b1:::bi:::bI . Using the inverted alignments in the maximum approximation, we obtain as search criterion: max I (p(JjI) max eI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) max bI 1 I Yi=1 [p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])) = = max I (p(JjI) max eI 1;bI 1 ( I Yi=1 p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) p(bijbi\u00f4\u0080\u0080\u00801; I; J) p(fbi jei)])); where the two products over i have been merged into a single product over i. p(eijei\u00f4\u0080\u0080\u00801 i\u00f4\u0080\u0080\u00802) is the trigram language model probability.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"For example, if is found in the list of person first names, the feature PersonFirstName is set to 1.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"If the token is the first word of a sentence, then this feature is set to 1.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"2.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"With each iteration more examples are assigned labels by both classifiers, while a high level of agreement (> 94%) is maintained between them.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Pseudo-labels are formed by taking seed labels on the labeled examples, and the output of the fixed classifier on the unlabeled examples.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"However, this optimistic search would not visit the entries necessary to store backoff information in the outgoing state.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In Chinese text, individual characters of the script, to which we shall refer by their traditional name of hanzi,Z are written one after another with no intervening spaces; a Chinese sentence is shown in Figure 1.3 Partly as a result of this, the notion \"word\" has never played a role in Chinese philological tradition, and the idea that Chinese lacks any\u00c2\u00ad thing analogous to words in European languages has been prevalent among Western sinologists; see DeFrancis (1984).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(3)), with one term for each classifier.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Syntactic decoders, such as cdec (Dyer et al., 2010), build state from null context then store it in the hypergraph node for later extension.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Let H be the set of hanzi, p be the set of pinyin syllables with tone marks, and P be the set of grammatical part-of-speech labels.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For the inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J), we drop the dependence on the target sentence length I. 2.2 Word Joining.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"It also incorporates the Good-Turing method (Baayen 1989; Church and Gale 1991) in estimating the likelihoods of previously unseen con\u00c2\u00ad structions, including morphological derivatives and personal names.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"25 16.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Our work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Because of this threshold, very few NE instance pairs could be used and hence the variety of phrases was also limited.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Second, we show that although the Penn Arabic Treebank is similar to other tree- banks in gross statistical terms, annotation consistency remains problematic.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"5 68.1 34.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"both MENE and IdentiFinder used more training data than we did (we used only the official MUC 6 and MUC7 training data).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"It is possible one could produce better models by introducing features describing constituents and their contexts because one parser could be much better than the majority of the others in particular situations.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In the case of, the most common usage is as an adverb with the pronunciation jiangl, so that variant is assigned the estimated cost of 5.98, and a high cost is assigned to nominal usage with the pronunciation jiang4.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For t = 1, T and for j = 1, 2: where 4 = exp(-jg'(xj,i)). practice, this greedy approach almost always results in an overall decrease in the value of Zco.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Le m\u00b4edicament de r\u00b4ef\u00b4erence de Silapo est EPREX\/ERYPO, qui contient de l\u2019\u00b4epo\u00b4etine alfa.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Using the virtual distribution Di (i) and pseudo-labels&quot;y.,\u201e values for Wo, W\u00b1 and W_ can be calculated for each possible weak hypothesis (i.e., for each feature x E Xi); the weak hypothesis with minimal value for Wo + 2\/WW _ can be chosen as before; and the weight for this weak hypothesis at = ln ww+411:) can be calculated.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In the last example, the less restrictive IbmS word reordering leads to a better translation, although the QmS translation is still acceptable.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"\"c' 0 + 0 \"0 ' \u00e2\u0080\u00a2 + a n t i g r e e d y x g r e e d y < > c u r r e n t m e t h o d o d i e t . o n l y \u00e2\u0080\u00a2 Taiwan 0 \u00c2\u00b7;; 0 c CD E i5 0\"' 9 9 \u00e2\u0080\u00a2 Mainland \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 \u00e2\u0080\u00a2 -0.30.20.1 0.0 0.1 0.2 Dimension 1 (62%) Figure 7 Classical metric multidimensional scaling of distance matrix, showing the two most significant dimensions.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For derived words that occur in our corpus we can estimate these costs as we would the costs for an underived dictionary entry.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This latter evaluation compares the performance of the system with that of several human judges since, as we shall show, even people do not agree on a single correct way to segment a text.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Topicalization of NP subjects in SVO configurations causes confusion with VO (pro-drop).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Eight out of the thirteen errors in the high frequency phrases in the CC-domain are the phrases in \u00e2\u0080\u009cagree\u00e2\u0080\u009d.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Since our objective is to compare distributions of bracketing discrepancies, we do not use heuristics to prune the set of nuclei.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Now we have sets of phrases which share a keyword and we have links between those sets.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The maximum precision row is the upper bound on accuracy if we could pick exactly the correct constituents from among the constituents suggested by the three parsers.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"For the purposes of EM, the &quot;observed&quot; data is {(xi, Ya\u2022 \u2022 \u2022 (xrn, Yrn), xfil, and the hidden data is {ym+i y}.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The correct resolution in sentence (c) depends on knowledge that kidnappers frequently blindfold their victims.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Figure 2 shows examples of lexical expectations that were learned for both domains.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"For monolingual treebank data we relied on the CoNLL-X and CoNLL-2007 shared tasks on dependency parsing (Buchholz and Marsi, 2006; Nivre et al., 2007).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In order to train this error-correction model, he divided his training corpus into 5 portions of 20% each.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, BABAR learned that agents that \u00e2\u0080\u009cassassinate\u00e2\u0080\u009d or \u00e2\u0080\u009cinvestigate a cause\u00e2\u0080\u009d are usually humans or groups (i.e., organizations).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Hence, we take the probability of the event fmnh analyzed as REL VB to be This means that we generate f and mnh independently depending on their corresponding PoS tags, and the context (as well as the syntactic relation between the two) is modeled via the derivation resulting in a sequence REL VB spanning the form fmnh. based on linear context.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The motivation for our more informal approach was the intuition that there are so many open problems in rhetorical analysis (and more so for German than for English; see below) that the main task is qualitative investigation, whereas rigorous quantitative analyses should be performed at a later stage.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"42 nator, the N31s can be measured well by counting, and we replace the expectation by the observation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"These systems are similar to those described by Pollard (1984) as Generalized Context-Free Grammars (GCFG's).","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"There, a lattice is used to represent the possible sentences resulting from an interpretation of an acoustic model.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Translation errors are reported in terms of multireference word error rate (mWER) and subjective sentence error rate (SSER).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Linguistic intuitions like those in the previous section inform language-specific annotation choices.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Our results suggest that current parsing models would benefit from better annotation consistency and enriched annotation in certain syntactic configurations.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"When this feature type was included, CoBoost chose this default feature at an early iteration, thereby giving non-abstaining pseudo-labels for all examples, with eventual convergence to the two classifiers agreeing by assigning the same label to almost all examples.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In such cases we assign all of the estimated probability mass to the form with the most likely pronunciation (determined by inspection), and assign a very small probability (a very high cost, arbitrarily chosen to be 40) to all other variants.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example: McCann initiated a new global system.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Its only purpose is 3 This follows since each \u00ce\u00b8t has St \u00e2\u0088\u0092 1 parameters and.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The developer explained that the loading process requires extra memory that it then frees. eBased on the ratio to SRI\u2019s speed reported in Guthrie and Hepple (2010) under different conditions.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"On the other hand, only very restricted reorderings are necessary, e.g. for the translation direction from Table 2: Coverage set hypothesis extensions for the IBM reordering.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(1997; 1999) do have some statistics that show how IdentiFinder performs when the training data is reduced.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"As far as we know, no other NERs have used information from the whole document (global) as well as information within the same sentence (local) in one framework.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"11 www.ling.unipotsdam.de\/sfb\/projekt a3.php 12 This step was carried out in the course of the diploma thesis work of David Reitter (2003), which de serves special mention here.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In this specific case, as these two titles could fill the same column of an IE table, we regarded them as paraphrases for the evaluation.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The first is an evaluation of the system's ability to mimic humans at the task of segmenting text into word-sized units; the second evaluates the proper-name identification; the third measures the performance on morphological analysis.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This best instance-weighting model beats the equivalant model without instance weights by between 0.6 BLEU and 1.8 BLEU, and beats the log-linear baseline by a large margin.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"of Tokens MENERGI 318 160,000 200 180,000 IdentiFinder \u00e2\u0080\u0093 650,000 \u00e2\u0080\u0093 790,000 MENE \u00e2\u0080\u0093 \u00e2\u0080\u0093 350 321,000 Table 4: Training Data MUC7 test accuracy.2 For MUC6, the reduction in error due to global features is 27%, and for MUC7,14%.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Unigrams also have 64-bit overhead for vocabulary lookup.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Ex: The regime gives itself the right...","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This is consistent with the nature of these two settings: log-linear combination, which effectively takes the intersection of IN and OUT, does relatively better on NIST, where the domains are broader and closer together.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Therefore, a populated probing hash table consists of an array of buckets that contain either one entry or are empty.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"On one end of the spectrum are clustering approaches that assign a single POS tag to each word type (Schutze, 1995; Lamar et al., 2010).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"However, it is also mostly political content (even if not focused on the internal workings of the European Union) and opinion.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(Blum and Mitchell 98) describe learning in the following situation: X = X1 X X2 where X1 and X2 correspond to two different &quot;views&quot; of an example.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Based on revision 4041, we modified Moses to print process statistics before terminating.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"While size of the resulting transducers may seem daunting-the segmenter described here, as it is used in the Bell Labs Mandarin TTS system has about 32,000 states and 209,000 arcs-recent work on minimization of weighted machines and transducers (cf.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"To formalize the approach, we introduce four verbgroup states S: Initial (I): A contiguous, initial block of source positions is covered.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Source sentence words are aligned with hypothesized target sentence words, where the choice of a new source word, which has not been aligned with a target word yet, is restricted1.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Projectivizing a dependency graph by lifting nonprojective arcs is a nondeterministic operation in the general case.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"(Thus the domain of the dev and test corpora matches IN.)","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"For a trigram language model, the partial hypotheses are of the form (e0; e; C; j).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We assume that M is in an existential state qA, with integers i1 and i2 representing zi in the (2i \u2014 1)th and 22th work tape, for 1 < i < k. For each rule p : A fp(B, C) such that fp is mapped onto the function fp defined by the following rule. jp((xi,.. \u2022 ,rnt), (1ii, \u2022 \u2022 \u2022 \u2022 Yn3))= (Zi , \u2022 \u2022 \u2022 , Zk) M breaks xi , zk into substrings xi, , xn, and yi,...,y&quot; conforming to the definition of fp.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"For other languages, we use the CoNLL-X multilingual dependency parsing shared task corpora (Buchholz and Marsi, 2006) which include gold POS tags (used for evaluation).","label":0},{"source_text":"There is no global pruning.","target_text":"There are 13 types of extensions needed to describe the verbgroup reordering.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"x EG, has agreed to be bought by H x EG, now owned by H x H to acquire EG x H\u00e2\u0080\u0099s agreement to buy EG Three of those phrases are actually paraphrases, but sometime there could be some noise; such as the second phrase above.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The results are given in Table 4.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In an existential state an ATM behaves like a nondeterministic TM, accepting if one of the applicable moves leads to acceptance; in an universal state the ATM accepts if all the applicable moves lead to acceptance.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Therefore, a populated probing hash table consists of an array of buckets that contain either one entry or are empty.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"3 68.9 50.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"BABAR uses the log-likelihood statistic (Dunning, 1993) to evaluate the strength of a co-occurrence relationship.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"There are two differences between this method and the DL-CoTrain algorithm: spelling and contextual features, alternating between labeling and learning with the two types of features.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Recently, this topic has been getting more attention, as is evident from the Paraphrase Workshops in 2003 and 2004, driven by the needs of various NLP applications.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Given n-gram counts {cn}Nn=1, we use Flog2 c1] bits per vocabulary identifier and Flog2 cn] per index into the table of ngrams.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The details are given in (Tillmann, 2000).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The Expectation Maximization (EM) algorithm (Dempster, Laird and Rubin 77) is a common approach for unsupervised training; in this section we describe its application to the named entity problem.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Finally, other packages implement language model estimation while we are currently dependent on them to generate an ARPA file.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The choice of the particular newspaper was motivated by the fact that the language used in a regional daily is somewhat simpler than that of papers read nationwide.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The iw all map variant uses a non-0 y weight on a uniform prior in p,,(s t), and outperforms a version with y = 0 (iw all) and the \u201cflattened\u201d variant described in section 3.2.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The first setting uses the European Medicines Agency (EMEA) corpus (Tiedemann, 2009) as IN, and the Europarl (EP) corpus (www.statmt.org\/europarl) as OUT, for English\/French translation in both directions.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"In particular, the decision to represent arguments in verb- initial clauses as VP internal makes VSO and VOS configurations difficult to distinguish.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Global features are extracted from other occurrences of the same token in the whole document.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For natural disasters, BABAR generated 20,479 resolutions: 11,652 from lexical seeding and 8,827 from syntactic seeding.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The probabilities are incorporated into the DempsterShafer model using Equation 1.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In our model there are no such hyper-parameters, and the performance is the result of truly joint disambiguation. sults.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"When translating the sentence monotonically from left to right, the translation of the German finite verb 'kann', which is the left verbal brace in this case, is postponed until the German noun phrase 'mein Kollege' is translated, which is the subject of the sentence.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Variants of alif are inconsistently used in Arabic texts.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Sorted arrays store key-value pairs in an array sorted by key, incurring no space overhead.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"If the log backoff of wnf is also zero (it may not be in filtered models), then wf should be omitted from the state.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The frequency of the Company \u00e2\u0080\u0093 Company domain ranks 11th with 35,567 examples.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Otherwise for the predecessor search hypothesis, we would have chosen a position that would not have been among the first n uncovered positions.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This logic applies recursively: if wnf+1 similarly does not extend and has zero log backoff, it too should be omitted, terminating with a possibly empty context.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We can now compare this algorithm to that of (Yarowsky 95).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Queries detect the invalid probability, using the node only if it leads to a longer match.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Note that we check for , the word preceding the consecutive sequence of initCaps tokens, since person prefixes like Mr., Dr., etc are not part of person names, whereas corporate suffixes like Corp., Inc., etc are part of corporate names.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"One of such approaches uses comparable documents, which are sets of documents whose content are found\/known to be almost the same, such as different newspaper stories about the same event [Shinyama and Sekine 03] or different translations of the same story [Barzilay 01].","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"BABAR achieved recall in the 4250% range for both domains, with 76% precision overall for terrorism and 87% precision for natural disasters.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"For all these annotation taks, G\u00c2\u00a8otze developed a series of questions (essentially a decision tree) designed to lead the annotator to the ap propriate judgement.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Among these are words derived by various productive processes, including: 1.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"In the rhetorical tree, nuclearity information is then used to extract a \u00e2\u0080\u009ckernel tree\u00e2\u0080\u009d that supposedly represents the key information from which the summary can be generated (which in turn may involve co-reference information, as we want to avoid dangling pronouns in a summary).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We now turn our attention to the recognition of string languages generated by these formalisms (LCFRL's).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(For some recent corpus-based work on Chinese abbreviations, see Huang, Ahrens, and Chen [1993].)","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"As the two NE categories are the same, we can\u00e2\u0080\u0099t differentiate phrases with different orders of par ticipants \u00e2\u0080\u0093 whether the buying company or the to-be-bought company comes first.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"As a result, Arabic sentences are usually long relative to English, especially after Length English (WSJ) Arabic (ATB) \u00e2\u0089\u00a4 20 41.9% 33.7% \u00e2\u0089\u00a4 40 92.4% 73.2% \u00e2\u0089\u00a4 63 99.7% 92.6% \u00e2\u0089\u00a4 70 99.9% 94.9% Table 2: Frequency distribution for sentence lengths in the WSJ (sections 2\u00e2\u0080\u009323) and the ATB (p1\u00e2\u0080\u00933).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We examine both the complexity of the paths of trees in the tree sets, and the kinds of dependencies that the formalisms can impose between paths.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We suspect that the higher precision in the disasters domain may be due to its substantially larger training corpus.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Email: gale@research.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"2.4 Underspecified rhetorical structure.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Experiments using data from the Prague Dependency Treebank show that the combined system can handle nonprojective constructions with a precision sufficient to yield a significant improvement in overall parsing accuracy.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Other approaches encode sparsity as a soft constraint.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Under certain conditions the constituent voting and na\u00efve Bayes constituent combination techniques are guaranteed to produce sets of constituents with no crossing brackets.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(See Sproat and Shih 1995.)","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The knowledge base then can be tested for its relation-inference capabilities on the basis of full-blown co-reference information.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This style of naming is never required and seems to be losing currency.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We will evaluate various specific aspects of the segmentation, as well as the overall segmentation per\u00c2\u00ad formance.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Note that in our model, conditioned on T , there is precisely one t which has nonzero probability for the token component, since for each word, exactly one \u00ce\u00b8t has support.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We hope that this will allow practitioners to apply our approach directly to languages for which no resources are available.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"However, we note that the final conditional estimates p(s|t) from a given phrase table maximize the likelihood of joint empirical phrase pair counts over a word-aligned corpus.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"3 61.7 38.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Language models are widely applied in natural language processing, and applications such as machine translation make very frequent queries.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Note that on some examples (around 2% of the test set) CoBoost abstained altogether; in these cases we labeled the test example with the baseline, organization, label.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Figure 1 shows sample sentences from these domains, which are widely divergent.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Morphological analyzers for Hebrew that analyze a surface form in isolation have been proposed by Segal (2000), Yona and Wintner (2005), and recently by the knowledge center for processing Hebrew (Itai et al., 2006).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Step 1.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The contextual role knowledge had the greatest impact on pronouns: +13% recall for terrorism and +15% recall for disasters, with a +1% precision gain in terrorism and a small precision drop of -3% in disasters.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Also, the argument has been made that machine translation performance should be evaluated via task-based evaluation metrics, i.e. how much it assists performing a useful task, such as supporting human translators or aiding the analysis of texts.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In this section, we brie y review our translation approach.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In these examples, the names identified by the two systems (if any) are underlined; the sentence with the correct segmentation is boxed.19 The differences in performance between the two systems relate directly to three issues, which can be seen as differences in the tuning of the models, rather than repre\u00c2\u00ad senting differences in the capabilities of the model per se.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"TPT has theoretically better locality because it stores ngrams near their suffixes, thereby placing reads for a single query in the same or adjacent pages.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"They cluster NE instance pairs based on the words in the contexts using a bag- of-words method.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"One of the strengths of the DempsterShafer model is its natural ability to recognize when several credible hypotheses are still in play.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Their results are then compared with the results of an automatic segmenter.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We use a patched version of BitPar allowing for direct input of probabilities instead of counts.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"One conclusion drawn from this annotation effort was that for humans and machines alike, 2 www.sfs.nphil.unituebingen.de\/Elwis\/stts\/ stts.html 3 www.coli.unisb.de\/sfb378\/negra-corpus\/annotate.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"We can do that . IbmS: Yes, wonderful.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"diesem 3.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Hence, there is a total of 29 classes (7 name classes 4 sub-classes 1 not-a-name class).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"2.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In Table 5 we present results from small test cor\u00c2\u00ad pora for the productive affixes handled by the current version of the system; as with names, the segmentation of morphologically derived words is generally either right or wrong.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"With a human evaluation we also showed that ATB inter-annotator agreement remains low relative to the WSJ corpus.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"If we remove this sample from the evaluation, then the ATB type-level error rises to only 37.4% while the n-gram error rate increases to 6.24%.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(Brin 98) ,describes a system for extracting (author, book-title) pairs from the World Wide Web using an approach that bootstraps from an initial seed set of examples.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The model was built with open vocabulary, modified Kneser-Ney smoothing, and default pruning settings that remove singletons of order 3 and higher.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A token that is allCaps will also be initCaps.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We will return to these issues in the discussion section.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For example, take ..., says Maury Cooper, a vice president at S.&P.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The family name set is restricted: there are a few hundred single-hanzi family names, and about ten double-hanzi ones.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Table 5: Effect of the beam threshold on the number of search errors (147 sentences).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"However, we have reason to doubt Chang et al.'s performance claims.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The pre terminal morphological analyses are mapped to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags provided with the tree- bank.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The PCFG was trained from the same sections of the Penn Treebank as the other three parsers.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"As in boosting, the algorithm works in rounds.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"If e < b then the key is not found.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"In addition to using the simple features directly, we also trained an SVM classifier with these features to distinguish between IN and OUT phrase pairs.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In Modern Hebrew (Hebrew), a Semitic language with very rich morphology, particles marking conjunctions, prepositions, complementizers and relativizers are bound elements prefixed to the word (Glinert, 1989).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We again assume a training set of n examples {x1 . xri} where the first m examples have labels {y1 ... yin}, and the last (n \u2014 m) examples are unlabeled.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Each parse is converted into a set of constituents represented as a tuples: (label, start, end).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"genitiveMark indicates recursive NPs with a indefinite nominal left daughter and an NP right daughter.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Machine Translation In this case my colleague can not visit on I n d i e s e m F a l l ka nn m e i n K o l l e g e a m the v i e r t e n M a i n i c h t b e s u c h e n S i e you fourth of May Figure 1: Reordering for the German verbgroup.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"From here on we will refer to the named-entity string itself as the spelling of the entity, and the contextual predicate as the context.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"There are thus some very good reasons why segmentation into words is an important task.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Unlike Kahane et al. (1998), we do not regard a projectivized representation as the final target of the parsing process.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"2.2 Syntactic structure.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Models To assess the marginal utility of each component of the model (see Section 3), we incremen- tally increase its sophistication.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Finally, the focus\/background partition is annotated, together with the focus question that elicits the corresponding answer.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"(1998) have also used a maximum entropy classifier that uses already tagged entities to help tag other entities.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"On MUC6 data, MENERGI also achieves performance comparable to IdentiFinder when trained on similar amount of training data.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"[Hasegawa et al. 04] reported only on relation discovery, but one could easily acquire para phrases from the results.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"This assumption, however, is not inherent to type-based tagging models.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We use MMAX for this annotation as well.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The number of top-ranked pairs to retain is chosen to optimize dev-set BLEU score.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Figure 3 Figure 1.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Because we don\u2019t have a separate development set, we used the training set to select among them and found 0.2 to work slightly better than 0.1 and 0.3.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Our first algorithm is similar to Yarowsky's, but with some important modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"each word in the lexicon whether or not each string is actually an instance of the word in question.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"2.5 Connectives with scopes.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Except for the left3The graphs satisfy all the well-formedness conditions given in section 2 except (possibly) connectedness.","label":0},{"source_text":"There is no global pruning.","target_text":"am 11.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Since our objective is to compare distributions of bracketing discrepancies, we do not use heuristics to prune the set of nuclei.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The increase is generally higher for PDT than for DDT, which indicates a greater diversity in non-projective constructions.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"One striking example is the error reduction for Spanish, which reduces error by 36.5% and 24.7% for the best and median settings respectively.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In addition to a heuristic based on decision list learning, we also presented a boosting-like framework that builds on ideas from (Blum and Mitchell 98).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"(If the TF\/IDF score of that word is below a threshold, the phrase is discarded.)","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In Table 7 we give results for several evaluation metrics.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Roughly speaking, the new algorithm presented in this paper performs a similar search, but instead minimizes a bound on the number of (unlabeled) examples on which two classifiers disagree.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We plan to explore more powerful techniques for exploiting the diversity of parsing methods.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Later, BerkeleyLM (Pauls and Klein, 2011) described ideas similar to ours.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Furthermore we expect the label distributions on the foreign to be fairly noisy, because the graph constraints have not been taken into account yet.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Suppose -y results from the adjunction of 71, ,-y, at the k distinct tree addresses 141, , nk in some elementary tree 7', respectively.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"As a result, Arabic sentences are usually long relative to English, especially after Length English (WSJ) Arabic (ATB) \u00e2\u0089\u00a4 20 41.9% 33.7% \u00e2\u0089\u00a4 40 92.4% 73.2% \u00e2\u0089\u00a4 63 99.7% 92.6% \u00e2\u0089\u00a4 70 99.9% 94.9% Table 2: Frequency distribution for sentence lengths in the WSJ (sections 2\u00e2\u0080\u009323) and the ATB (p1\u00e2\u0080\u00933).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Not surprisingly some semantic classes are better for names than others: in our corpora, many names are picked from the GRASS class but very few from the SICKNESS class.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This material is based upon work supported by the National Science Foundation Graduate Research Fellowship under Grant No.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We examine both the complexity of the paths of trees in the tree sets, and the kinds of dependencies that the formalisms can impose between paths.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"For example, the independence assumptions mean that the model fails to capture the dependence between specific and more general features (for example the fact that the feature full.-string=New_York is always seen with the features contains (New) and The baseline method tags all entities as the most frequent class type (organization). contains (York) and is never seen with a feature such as contains (Group) ).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The results in Table 1 show PROBING is 81% faster than TRIE, which is in turn 31% faster than the fastest baseline.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Not only is concatenation of these strings possible, but head wrapping can be used to split a string and wrap it around another string.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Storing state therefore becomes a time-space tradeoff; for example, we store state with partial hypotheses in Moses but not with each phrase.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"diesem 3.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The transition from f, to a final state transduces c to the grammatical tag PL with cost cost(unseen(f,)): cost(i\u00c2\u00a5JJ1l.ir,) == cost(i\u00c2\u00a5JJ1l.)","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We again assume a training set of n examples {x1 . xri} where the first m examples have labels {y1 ... yin}, and the last (n \u2014 m) examples are unlabeled.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This remains as future work.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We used the TF\/ITF metric to identify keywords.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The development set contained 44088 constituents in 2416 sentences and the test set contained 30691 constituents in 1699 sentences.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Another option is the closedsource data structures from Sheffield (Guthrie and Hepple, 2010).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The input to AdaBoost is a set of training examples ((xi , yi), , (x\u201e.\u201e yrn)).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In this section we present a partial evaluation of the current system, in three parts.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Three state-of-the-art statistical parsers are combined to produce more accurate parses, as well as new bounds on achievable Treebank parsing accuracy.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"This solution also obviates the need to perform word sense disambiguation.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, the company may refer to Company X in one paragraph and Company Y in another.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"2.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Such tag sequences are often treated as \u201ccomplex tags\u201d (e.g.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Instead, we resort to an iterative update based method.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In a more recent study than Chang et al., Wang, Li, and Chang (1992) propose a surname-driven, non-stochastic, rule-based system for identifying personal names.17 Wang, Li, and Chang also compare their performance with Chang et al.'s system.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This is appropriate in cases where it is sanctioned by Bayes\u2019 law, such as multiplying LM and TM probabilities, but for adaptation a more suitable framework is often a mixture model in which each event may be generated from some domain.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Figure 4 shows a constituent headed by a process nominal with an embedded adjective phrase.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"irL as the product of the probability estimate for i\u00c2\u00a5JJ1l., and the probability estimate just derived for unseen plurals in ir,: p(i\u00c2\u00a51J1l.ir,) p(i\u00c2\u00a51J1l.)p(unseen(f,)).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"With regard to purely morphological phenomena, certain processes are not han\u00c2\u00ad dled elegantly within the current framework Any process involving reduplication, for instance, does not lend itself to modeling by finite-state techniques, since there is no way that finite-state networks can directly implement the copying operations required.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Fortunately, performance was stable across various values, and we were able to use the same hyperparameters for all languages.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Further, Maamouri and Bies (2004) argued that the English guidelines generalize well to other languages.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Given an anaphor and candidate, BABAR checks (1) whether the semantic classes of the anaphor intersect with the semantic expectations of the caseframe that extracts the candidate, and (2) whether the semantic classes of the candidate intersect with the semantic ex pectations of the caseframe that extracts the anaphor.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"In Modern Hebrew (Hebrew), a Semitic language with very rich morphology, particles marking conjunctions, prepositions, complementizers and relativizers are bound elements prefixed to the word (Glinert, 1989).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In contrast, even though context can be helpful for resolving definite NPs, context can be trumped by the semantics of the nouns themselves.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"For humans, this characteristic can impede the acquisition of literacy.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Our experiments all concern the analytical annotation, and the first experiment is based only on the training part.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In contrast, NNP (proper nouns) form a large portion of vocabulary.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Subsets C of increasing cardinality c are processed.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The Wang, Li, and Chang system fails on fragment (b) because their system lacks the word youlyoul 'soberly' and misinterpreted the thus isolated first youl as being the final hanzi of the preceding name; similarly our system failed in fragment (h) since it is missing the abbreviation i:lJI!","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For each resolution in the training data, BABAR also associates the co-referring expression of an NP with the NP\u00e2\u0080\u0099s caseframe.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Only tokens with initCaps not found in commonWords are tested against each list in Table 2.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"One of the difficulties in Natural Language Processing is the fact that there are many ways to express the same thing or event.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"63 81.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Thus at each iteration the method induces at most n x k rules, where k is the number of possible labels (k = 3 in the experiments in this paper). step 3.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Next, we represent the input sentence as an unweighted finite-state acceptor (FSA) I over H. Let us assume the existence of a function Id, which takes as input an FSA A, and produces as output a transducer that maps all and only the strings of symbols accepted by A to themselves (Kaplan and Kay 1994).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"12 For English, our Evalb implementation is identical to the most recent reference (EVALB20080701).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"KenLM: Faster and Smaller Language Model Queries","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This is especially effective at reducing load time, since raw bytes are read directly to memory\u2014or, as happens with repeatedly used models, are already in the disk cache.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"As noted for the perplexity task, we do not expect cache to grow substantially with model size, so RandLM remains a low-memory option.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In this section we present a partial evaluation of the current system, in three parts.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"7 Big 5 is the most popular Chinese character coding standard in use in Taiwan and Hong Kong.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Frequency is calculated by counting the number of distinct previous tokens that each token has (e.g., if Electric Corp. is seen 3 times, and Manufacturing Corp. is seen 5 times during training, and Corp. is not seen with any other preceding tokens, then the \u00e2\u0080\u009cfrequency\u00e2\u0080\u009d of Corp. is 2).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Arabic is a morphologically rich language with a root-and-pattern system similar to other Semitic languages.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We first define &quot;pseudo-labels&quot;,-yt, as follows: = Yi t sign(g 0\\ 2\u2014 kx2,m < i < n Thus the first m labels are simply copied from the labeled examples, while the remaining (n \u2014 m) examples are taken as the current output of the second classifier.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The availability of these resources guided our selection of foreign languages.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"ICOC and CSPP contributed the greatest im provements.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"A possible probabilistic model for assigning probabilities to complex analyses of a surface form may be and indeed recent sequential disambiguation models for Hebrew (Adler and Elhadad, 2006) and Arabic (Smith et al., 2005) present similar models.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We have developed a coreference resolver called BABAR that uses contextual role knowledge to make coreference decisions.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The final score is obtained from: max e;e0 j2fJ\u00f4\u0080\u0080\u0080L;;Jg p($je; e0) Qe0 (e; I; f1; ; Jg; j); where p($je; e0) denotes the trigram language model, which predicts the sentence boundary $ at the end of the target sentence.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"For effectively annotating connectives\/scopes, we found that existing annotation tools were not well-suited, for two reasons: \u00e2\u0080\u00a2 Some tools are dedicated to modes of annotation (e.g., tiers), which could only quite un-intuitively be used for connectives and scopes.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"There is twice as much language modelling data, since training data for the machine translation system is filtered against sentences of length larger than 40 words.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We will also directly compare with a baseline similar to the Matsoukas et al approach in order to measure the benefit from weighting phrase pairs (or ngrams) rather than full sentences.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Multiple features can be used for the same token.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Having defined LCFRS's, in Section 4.2 we established the semilinearity (and hence constant growth property) of the languages generated.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, for our purposes it is not sufficient to repre\u00c2\u00ad sent the morphological decomposition of, say, plural nouns: we also need an estimate of the cost of the resulting word.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The DempsterShafer rule for combining pdfs is: to {C}, meaning that it is 70% sure the correct hypothesis is C. The intersection of these sets is the null set because these beliefs are contradictory.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"AdaBoost was first introduced in (Freund and Schapire 97); (Schapire and Singer 98) gave a generalization of AdaBoost which we will use in this paper.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This is in general very difficult, given the extremely free manner in which Chinese given names are formed, and given that in these cases we lack even a family name to give the model confidence that it is identifying a name.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The development of automatic scoring methods is an open field of research.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Certainly these linguistic factors increase the difficulty of syntactic disambiguation.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In particular, it may not be possible to learn functions fi (x f2(x2,t) for i = m + 1...n: either because there is some noise in the data, or because it is just not realistic to expect to learn perfect classifiers given the features used for representation.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Given the closeness of most systems and the wide over-lapping confidence intervals it is hard to make strong statements about the correlation between human judgements and automatic scoring methods such as BLEU.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"5 Related Work.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The iw all map variant uses a non-0 y weight on a uniform prior in p,,(s t), and outperforms a version with y = 0 (iw all) and the \u201cflattened\u201d variant described in section 3.2.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"To be a member of LCFRS a formalism must satisfy two restrictions.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"While the semantic aspect of radicals is by no means completely predictive, the semantic homogeneity of many classes is quite striking: for example 254 out of the 263 examples (97%) of the INSECT class listed by Wieger (1965, 77376) denote crawling or invertebrate animals; similarly 21 out of the 22 examples (95%) of the GHOST class (page 808) denote ghosts or spirits.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"It is sometimes claimed that one of the advantages of dependency grammar over approaches based on constituency is that it allows a more adequate treatment of languages with variable word order, where discontinuous syntactic constructions are more common than in languages like English (Mel\u2019\u02c7cuk, 1988; Covington, 1990).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"An additional case of super-segmental morphology is the case of Pronominal Clitics.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold, 30.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We use a squared loss to penalize neighboring vertices that have different label distributions: kqi \u2212 qjk2 = Ey(qi(y) \u2212 qj(y))2, and additionally regularize the label distributions towards the uniform distribution U over all possible labels Y.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"So, who won the competition?","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"However, 1 http:\/\/maxent.sourceforge.net 3.2 Testing.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The final model tions.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Pseudo-code describing the generalized boosting algorithm of Schapire and Singer is given in Figure 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Machine Translation In this case my colleague can not visit on I n d i e s e m F a l l ka nn m e i n K o l l e g e a m the v i e r t e n M a i n i c h t b e s u c h e n S i e you fourth of May Figure 1: Reordering for the German verbgroup.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In section 4 we evaluate these transformations with respect to projectivized dependency treebanks, and in section 5 they are applied to parser output.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Many packages perform language model queries.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We received submissions from 14 groups from 11 institutions, as listed in Figure 2.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As we shall argue, the semantic class affiliation of a hanzi constitutes useful information in predicting its properties.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"lla\/llb and 14a\/14b respectively).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We used 22 features for the logistic weighting model, divided into two groups: one intended to reflect the degree to which a phrase pair belongs to general language, and one intended to capture similarity to the IN domain.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"More recently, Subramanya et al. (2010) defined a graph over the cliques in an underlying structured prediction model.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"When the connective is an adverbial, there is much less clarity as to the range of the spans.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Consequently, all three parsers prefer the nominal reading.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"2 Chinese ?l* han4zi4 'Chinese character'; this is the same word as Japanese kanji..","label":0},{"source_text":"Here we present two algorithms.","target_text":"In principle a feature could be an arbitrary predicate of the (spelling, context) pair; for reasons that will become clear, features are limited to querying either the spelling or context alone.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"an event.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In the numerator, however, the counts of ni1s are quite irregular, in\u00c2\u00ad cluding several zeros (e.g., RAT, none of whose members were seen).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The choice of the particular newspaper was motivated by the fact that the language used in a regional daily is somewhat simpler than that of papers read nationwide.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The work reported here is closely related to [Ha- segawa et al. 04].","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Memory usage is likely much lower than ours. fThe original paper (Germann et al., 2009) provided only 2s of query timing and compared with SRI when it exceeded available RAM.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The paper explains the design decisions taken in the annotations, and describes a number of applications using this corpus with its multi-layer annotation.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We hope that this will allow practitioners to apply our approach directly to languages for which no resources are available.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Furthermore, the systematic way in which particles are prefixed to one another and onto an open-class category gives rise to a distinct sort of morphological ambiguity: space-delimited tokens may be ambiguous between several different segmentation possibilities.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"This scenario is applicable to a large set of languages and has been considered by a number of authors in the past (Alshawi et al., 2000; Xi and Hwa, 2005; Ganchev et al., 2009).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"In these experiments, the input lacks segmentation markers, hence the slightly different dev set baseline than in Table 6.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Because conjunctions are elevated in the parse trees when they separate recursive constituents, we choose the right sister instead of the category of the next word.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The secondary classifier in (Borthwick, 1999) uses information not just from the current article, but also from the whole test corpus, with an additional feature that indicates if the information comes from the same document or from another document.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"2 62.2 39.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This process is repeated 5 times by rotating the data appropriately.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Given that part-of-speech labels are properties of words rather than morphemes, it follows that one cannot do part-of-speech assignment without having access to word-boundary information.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"92 76.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Part of the gap between resident and virtual memory is due to the time at which data was collected.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Then, the German infinitive 'besuchen' and the negation particle 'nicht' are translated.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"1,000 of these were picked at random, and labeled by hand to produce a test set.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Semantic expectations are analogous to lexical expectations except that they represent semantic classes rather than nouns.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Also surprising is the low test set OOV rate given the possibility of morphological variation in Arabic.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Across all languages, high performance can be attained by selecting a single tag per word type.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We indicate whether a context with zero log backoff will extend using the sign bit: +0.0 for contexts that extend and \u22120.0 for contexts that do not extend.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The first concerns how to deal with ambiguities in segmentation.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Handling Unknown tokens When handling unknown tokens in a language such as Hebrew various important aspects have to be borne in mind.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Instead, we extend the variation n-gram method of Dickinson (2005) to compare annotation error rates in the WSJ and ATB.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A minimal requirement for building a Chinese word segmenter is obviously a dictionary; furthermore, as has been argued persuasively by Fung and Wu (1994), one will perform much better at segmenting text by using a dictionary constructed with text of the same genre as the text to be segmented.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Thatcher (1973) describes a tee pumping lemma for recognizable sets related to the string pumping lemma for regular sets.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"To date we have not done a separate evaluation of foreign-name recognition.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"3.1 Corpora.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-8) (1-6) lcc (1-6) (1-7) (1-4) utd (1-7) (1-6) (2-7) upc-mr (1-8) (1-6) (1-7) nrc (1-7) (2-6) (8) ntt (1-8) (2-8) (1-7) cmu (3-7) (4-8) (2-7) rali (5-8) (3-9) (3-7) systran (9) (8-9) (10) upv (10) (10) (9) Spanish-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-7) (1-6) (1-5) ntt (1-7) (1-8) (1-5) lcc (1-8) (2-8) (1-4) utd (1-8) (2-7) (1-5) nrc (2-8) (1-9) (6) upc-mr (1-8) (1-6) (7) uedin-birch (1-8) (2-10) (8) rali (3-9) (3-9) (2-5) upc-jg (7-9) (6-9) (9) upv (10) (9-10) (10) German-English (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) uedin-phi (1-2) (1) (1) lcc (2-7) (2-7) (2) nrc (2-7) (2-6) (5-7) utd (3-7) (2-8) (3-4) ntt (2-9) (2-8) (3-4) upc-mr (3-9) (6-9) (8) rali (4-9) (3-9) (5-7) upc-jmc (2-9) (3-9) (5-7) systran (3-9) (3-9) (10) upv (10) (10) (9) Figure 7: Evaluation of translation to English on in-domain test data 112 English-French (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) nrc (1-5) (1-5) (1-6) upc-mr (1-4) (1-5) (1-6) upc-jmc (1-6) (1-6) (1-5) systran (2-7) (1-6) (7) utd (3-7) (3-7) (3-6) rali (1-7) (2-7) (1-6) ntt (4-7) (4-7) (1-5) English-Spanish (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) ms (1-5) (1-7) (7-8) upc-mr (1-4) (1-5) (1-4) utd (1-5) (1-6) (1-4) nrc (2-7) (1-6) (5-6) ntt (3-7) (1-6) (1-4) upc-jmc (2-7) (2-7) (1-4) rali (5-8) (6-8) (5-6) uedin-birch (6-9) (6-10) (7-8) upc-jg (9) (8-10) (9) upv (9-10) (8-10) (10) English-German (In Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-5) (3-5) ntt (1-5) (2-6) (1-3) upc-jmc (1-5) (1-4) (1-3) nrc (2-4) (1-5) (4-5) rali (3-6) (2-6) (1-4) systran (5-6) (3-6) (7) upv (7) (7) (6) Figure 8: Evaluation of translation from English on in-domain test data 113 French-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-5) (1-8) (1-4) cmu (1-8) (1-9) (4-7) systran (1-8) (1-7) (9) lcc (1-9) (1-9) (1-5) upc-mr (2-8) (1-7) (1-3) utd (1-9) (1-8) (3-7) ntt (3-9) (1-9) (3-7) nrc (3-8) (3-9) (3-7) rali (4-9) (5-9) (8) upv (10) (10) (10) Spanish-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-jmc (1-2) (1-6) (1-3) uedin-birch (1-7) (1-6) (5-8) nrc (2-8) (1-8) (5-7) ntt (2-7) (2-6) (3-4) upc-mr (2-8) (1-7) (5-8) lcc (4-9) (3-7) (1-4) utd (2-9) (2-8) (1-3) upc-jg (4-9) (7-9) (9) rali (4-9) (6-9) (6-8) upv (10) (10) (10) German-English (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1-4) (1-4) (7-9) uedin-phi (1-6) (1-7) (1) lcc (1-6) (1-7) (2-3) utd (2-7) (2-6) (4-6) ntt (1-9) (1-7) (3-5) nrc (3-8) (2-8) (7-8) upc-mr (4-8) (6-8) (4-6) upc-jmc (4-8) (3-9) (2-5) rali (8-9) (8-9) (8-9) upv (10) (10) (10) Figure 9: Evaluation of translation to English on out-of-domain test data 114 English-French (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1) (1) upc-jmc (2-5) (2-4) (2-6) upc-mr (2-4) (2-4) (2-6) utd (2-6) (2-6) (7) rali (4-7) (5-7) (2-6) nrc (4-7) (4-7) (2-5) ntt (4-7) (4-7) (3-6) English-Spanish (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) upc-mr (1-3) (1-6) (1-2) ms (1-7) (1-8) (6-7) utd (2-6) (1-7) (3-5) nrc (1-6) (2-7) (3-5) upc-jmc (2-7) (1-6) (3-5) ntt (2-7) (1-7) (1-2) rali (6-8) (4-8) (6-8) uedin-birch (6-10) (5-9) (7-8) upc-jg (8-9) (9-10) (9) upv (9) (8-9) (10) English-German (Out of Domain) Adequacy (rank) Fluency (rank) BLEU (rank) systran (1) (1-2) (1-6) upc-mr (2-3) (1-3) (1-5) upc-jmc (2-3) (3-6) (1-6) rali (4-6) (4-6) (1-6) nrc (4-6) (2-6) (2-6) ntt (4-6) (3-5) (1-6) upv (7) (7) (7) Figure 10: Evaluation of translation from English on out-of-domain test data 115 French-English In domain Out of Domain Adequacy Adequacy 0.3 0.3 \u2022 0.2 0.2 0.1 0.1 -0.0 -0.0 -0.1 -0.1 -0.2 -0.2 -0.3 -0.3 -0.4 -0.4 -0.5 -0.5 -0.6 -0.6 -0.7 -0.7 \u2022upv -0.8 -0.8 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 \u2022upv \u2022systran upcntt \u2022 rali upc-jmc \u2022 cc Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022systran \u2022upv upc -jmc \u2022 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022 \u2022 \u2022 td t cc upc- \u2022 rali 21 22 23 24 25 26 27 28 29 30 31 15 16 17 18 19 20 21 22 Figure 11: Correlation between manual and automatic scores for French-English 116 Spanish-English Figure 12: Correlation between manual and automatic scores for Spanish-English -0.3 -0.4 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 \u2022upv -0.4 \u2022upv -0.3 In Domain \u2022upc-jg Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 Out of Domain \u2022upc-jmc \u2022nrc \u2022ntt Adequacy upc-jmc \u2022 \u2022 \u2022lcc \u2022 rali \u2022 \u2022rali -0.7 -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 \u2022 \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 ntt \u2022 upc-mr \u2022lcc \u2022utd \u2022upc-jg \u2022rali Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upc-jmc \u2022 uedin-birch -0.5 -0.5 \u2022upv 23 24 25 26 27 28 29 30 31 32 19 20 21 22 23 24 25 26 27 28 117 In Domain Out of Domain Adequacy Adequacy German-English 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 lcc \u2022 upc-jmc \u2022systran \u2022upv Fluency \u2022ula \u2022upc-mr \u2022lcc 15 16 17 18 19 20 21 22 23 24 25 26 27 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022systran \u2022upv \u2022uedin-phi -jmc \u2022rali \u2022systran -0.3 -0.4 -0.5 -0.6 \u2022upv 12 13 14 15 16 17 18 19 20 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 Fluency uedin-phi \u2022 \u2022 \u2022utd \u2022upc-jmc \u2022upc-mr 0.4 \u2022rali -0.3 -0.4 -0.5 \u2022upv 12 13 14 15 16 17 18 19 20 0.3 0.2 0.1 -0.0 -0.1 -0.2 English-French In Domain Out of Domain Adequacy Adequacy .","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We refer to different readings as different analyses whereby the segments are deterministic given the sequence of PoS tags.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"In future work we plan to try this approach with more competitive SMT systems, and to extend instance weighting to other standard SMT components such as the LM, lexical phrase weights, and lexicalized distortion.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Finally, other packages implement language model estimation while we are currently dependent on them to generate an ARPA file.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"First, it directly encodes linguistic intuitions about POS tag assignments: the model structure reflects the one-tag-per-word property, and a type- level tag prior captures the skew on tag assignments (e.g., there are fewer unique determiners than unique nouns).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The sequence of states needed to carry out the word reordering example in Fig.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"As Figure 1 shows, this word has no high-confidence alignment in the Italian-English bitext.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The use of the Good-Turing equation presumes suitable estimates of the unknown expectations it requires.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"To initialize the graph for label propagation we use a supervised English tagger to label the English side of the bitext.7 We then simply count the individual labels of the English tokens and normalize the counts to produce tag distributions over English word types.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"An error count of 0:0 is assigned to a perfect translation, and an error count of 1:0 is assigned to a semantically and syntactically wrong translation.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Several such particles may be prefixed onto a single stem, in which case the affixation is subject to strict linear precedence constraints.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"On the other hand, if it is seen as McCann Pte.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Blum and Mitchell 98) describe learning in the following situation: X = X1 X X2 where X1 and X2 correspond to two different &quot;views&quot; of an example.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"To evaluate proper-name identification, we randomly se\u00c2\u00ad lected 186 sentences containing 12,000 hanzi from our test corpus and segmented the text automatically, tagging personal names; note that for names, there is always a sin\u00c2\u00ad gle unambiguous answer, unlike the more general question of which segmentation is correct.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The type-level tag assignments T generate features associated with word types W . The tag assignments constrain the HMM emission parameters \u00ce\u00b8.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Section 5 covers relevant previous work on SMT adaptation, and section 6 concludes.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"tai2du2 'Taiwan Independence.'","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Informally, two or more paths can be dependent on each other: for example, they could be required to be of equal length as in the trees in Figure 4. generates such a tree set.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"In other words, the set of hidden states F was chosen to be the fine set of treebank tags.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Unexpectedly, because the ministries of treasury and education both had prepared the teacher plan together.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"pronunciations of individual words; they also need to compute intonational phrase boundaries in long utterances and assign relative prominence to words in those utterances.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We can better predict the probability of an unseen hanzi occurring in a name by computing a within-class Good-Turing estimate for each radical class.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In TAG's the elementary tree and addresses where adjunction takes place are used to instantiate the operation.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"(1999) did not report using any dictionaries, but mentioned in a footnote that they have added list membership features, which have helped marginally in certain domains.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We are currently exploring other methods that employ similar ideas and their formal properties.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The combining technique must act as a multi-position switch indicating which parser should be trusted for the particular sentence.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"IdentiFinder ' 99' s results are considerably better than IdentiFinder ' 97' s. IdentiFinder' s performance in MUC7 is published in (Miller et al., 1998).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We do not experiment with models larger than physical memory in this paper because TPT is unreleased, factors such as disk speed are hard to replicate, and in such situations we recommend switching to a more compact representation, such as RandLM.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We report micro-averaged (whole corpus) and macro-averaged (per sentence) scores along add a constraint on the removal of punctuation, which has a single tag (PUNC) in the ATB.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Our smoothing procedure takes into account all the aforementioned aspects and works as follows.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"We have shown that the maximum entropy framework is able to use global information directly.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"IRSTLM\u2019s quantized variant is the inspiration for our quantized variant.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Standard SMT systems have a hierarchical parameter structure: top-level log-linear weights are used to combine a small set of complex features, interpreted as log probabilities, many of which have their own internal parameters and objectives.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Of these cases, 38 were temporal expressions (either a day of the week or month of the year).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This is less effective in our setting, where IN and OUT are disparate.","label":0},{"source_text":"There is no global pruning.","target_text":"Machine Translation In this case my colleague can not visit on I n d i e s e m F a l l ka nn m e i n K o l l e g e a m the v i e r t e n M a i n i c h t b e s u c h e n S i e you fourth of May Figure 1: Reordering for the German verbgroup.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"This approach needs a phrase as an initial seed and thus the possible relationships to be extracted are naturally limited.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"This stage of label propagation results in a tag distribution ri over labels y, which encodes the proportion of times the middle word of ui E Vf aligns to English words vy tagged with label y: The second stage consists of running traditional label propagation to propagate labels from these peripheral vertices Vf\ufffd to all foreign language vertices in the graph, optimizing the following objective: 5 POS Induction After running label propagation (LP), we compute tag probabilities for foreign word types x by marginalizing the POS tag distributions of foreign trigrams ui = x\u2212 x x+ over the left and right context words: where the qi (i = 1, ... , |Vf|) are the label distributions over the foreign language vertices and \u00b5 and \u03bd are hyperparameters that we discuss in \u00a76.4.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Since pronouns carry little semantics of their own, resolving them depends almost entirely on context.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"With a human evaluation we also showed that ATB inter-annotator agreement remains low relative to the WSJ corpus.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"9 50.2 +P RI OR be st me dia n 47.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Table 1 shows results of the benchmark.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In particular, the decision to represent arguments in verb- initial clauses as VP internal makes VSO and VOS configurations difficult to distinguish.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"These results are promising and there are several avenues for improving on these results.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Otherwise, the scope of the search problem shrinks recursively: if A[pivot] < k then this becomes the new lower bound: l +\u2014 pivot; if A[pivot] > k then u +\u2014 pivot.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Because these two words have identical complements, syntax rules are typically unhelpful for distinguishing between them.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The ATB annotation distinguishes between verbal and nominal readings of maSdar process nominals.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"For example, from the sentence \u00e2\u0080\u009cMr.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"4.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"However, there are phrases which express the same meanings even though they do not share the same keyword.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The knowledge base then can be tested for its relation-inference capabilities on the basis of full-blown co-reference information.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Email: gale@research.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Almost all annotators expressed their preference to move to a ranking-based evaluation in the future.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Inflectional features marking pronominal elements may be attached to different kinds of categories marking their pronominal complements.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Two common cases are the attribu tive adjective and the process nominal _; maSdar, which can have a verbal reading.4 At tributive adjectives are hard because they are or- thographically identical to nominals; they are inflected for gender, number, case, and definiteness.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":".","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We have developed a coreference resolver called BABAR that uses contextual role knowledge to make coreference decisions.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Our experiments all concern the analytical annotation, and the first experiment is based only on the training part.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We attempt to formalize this notion in terms of the tee pumping lemma which can be used to show that a tee set does not have dependent paths.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We evaluate our approach on seven languages: English, Danish, Dutch, German, Portuguese, Spanish, and Swedish.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"2.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The discovered paraphrases can be a big help to reduce human labor and create a more comprehensive pattern set.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"We apply a beam search concept as in speech recognition.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"If the context wnf will never extend to the right (i.e. wnf v is not present in the model for all words v) then no subsequent query will match the full context.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"In future work we plan to try this approach with more competitive SMT systems, and to extend instance weighting to other standard SMT components such as the LM, lexical phrase weights, and lexicalized distortion.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The performance was 80.99% recall and 61.83% precision.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"an event.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Various segmentation approaches were then compared with human performance: 1.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We also collapse unary chains withidentical basic categories like NP \u00e2\u0086\u0092 NP.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We were intentionally lenient with our baselines: bilingual information by projecting POS tags directly across alignments in the parallel data.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Applications The discovered paraphrases have multiple applications.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"98 15.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(1998) have also used a maximum entropy classifier that uses already tagged entities to help tag other entities.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We tabulate this increase in Table 3.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"A morpheme, on the other hand, usually corresponds to a unique hanzi, though there are a few cases where variant forms are found.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"These three parsers have given the best reported parsing results on the Penn Treebank Wall Street Journal corpus (Marcus et al., 1993).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"This allow the learners to &quot;bootstrap&quot; each other by filling the labels of the instances on which the other side has abstained so far.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Next we show that the ATB is similar to other tree- banks in gross statistical terms, but that annotation consistency remains low relative to English (\u00c2\u00a73).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"As noted for the perplexity task, we do not expect cache to grow substantially with model size, so RandLM remains a low-memory option.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The use of weighted transducers in particular has the attractive property that the model, as it stands, can be straightforwardly interfaced to other modules of a larger speech or natural language system: presumably one does not want to segment Chinese text for its own sake but instead with a larger purpose in mind.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The final block in table 2 shows models trained on feature subsets and on the SVM feature described in 3.4.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"9 61.0 44.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Each vertex within a connected component must have the same label \u2014 in the binary classification case, we need a single labeled example to identify which component should get which label.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In contrast to these approaches, our method directly incorporates these constraints into the structure of the model.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"A total of 13,976 phrases were grouped.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, in MUC6, there are four zones (TXT, HL, DATELINE, DD).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"In Figure 4 we show an example of variation between the parsing models.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This implies, therefore, that a major factor in the performance of a Chinese segmenter is the quality of the base dictionary, and this is probably a more important factor-from the point of view of performance alone-than the particular computational methods used.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Again, the idea is that having a picture of syntax, co-reference, and sentence-internal information structure at one\u00e2\u0080\u0099s disposal should aid in finding models of discourse structure that are more explanatory and can be empirically supported.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The NP is a complement to a preposition, which is the head of a PP.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"(2) Description Source Location Names http:\/\/www.timeanddate.com http:\/\/www.cityguide.travel-guides.com http:\/\/www.worldtravelguide.net Corporate Names http:\/\/www.fmlx.com Person First Names http:\/\/www.census.gov\/genealogy\/names Person Last Names Table 2: Sources of Dictionaries The McCann family . . .","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Initially, the DempsterShafer model assumes that all hypotheses are equally likely, so it creates a set called \u00ce\u00b8 that includes all hypotheses.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Gabbard and Kulick (2008) show that there is significant attachment ambiguity associated with iDafa, which occurs in 84.3% of the trees in our development set.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"A greedy algorithm (or maximum-matching algorithm), GR: proceed through the sentence, taking the longest match with a dictionary entry at each point.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"As indicated in Figure 1(c), apart from this correct analysis, there is also the analysis taking B ri4 as a word (e.g., a common abbreviation for Japan), along with X:\u00c2\u00a5 wen2zhangl 'essay\/ and f!!.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"They return a value in the range [0,1], where 0 indicates neutrality and 1 indicates the strongest belief that the candidate and anaphor are coreferent.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"On the other hand, in a translation system one probably wants to treat this string as a single dictionary word since it has a conventional and somewhat unpredictable translation into English.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"For the error counts, a range from 0:0 to 1:0 is used.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Local features are features that are based on neighboring tokens, as well as the token itself.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In defining LCFRS's, we hope to generalize the definition of CFG's to formalisms manipulating any structure, e.g. strings, trees, or graphs.","label":0},{"source_text":"There is no global pruning.","target_text":"Mai.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The implication of this ambiguity for a parser is that the yield of syntactic trees no longer consists of spacedelimited tokens, and the expected number of leaves in the syntactic analysis in not known in advance.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The NP is a complement to a preposition, which is the head of a PP.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"By aggregating the POS labels of the English tokens to types, we can generate label distributions for the English vertices.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The procedure using the tagged sentences to discover paraphrases takes about one hour on a 2GHz Pentium 4 PC with 1GB of memory.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Table 9: Dev set results for sentences of length \u00e2\u0089\u00a4 70.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The types of patterns produced by AutoSlog are outlined in (Riloff, 1996).","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"A number of PCC commentaries will be read by professional news speakers and prosodic features be annotated, so that the various annotation layers can be set into correspondence with intonation patterns.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Although these existential NPs do not need a prior referent, they may occur multiple times in a document.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Sie.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"It is worth noting that the middle words of the Italian trigrams are nouns too, which exhibits the fact that the similarity metric connects types having the same syntactic category.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Finally, we incorporate the instance-weighting model into a general linear combination, and learn weights and mixing parameters simultaneously. where c\u03bb(s, t) is a modified count for pair (s, t) in OUT, u(s|t) is a prior distribution, and y is a prior weight.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The probabilistic version of this procedure is straightforward: We once again assume independence among our various member parsers.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We used the MUC4 terrorism corpus (MUC4 Proceedings, 1992) and news articles from the Reuter\u00e2\u0080\u0099s text collection8 that had a subject code corresponding to natural disasters.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For IE, the system must be able to distinguish between semantically similar noun phrases that play different roles in an event.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"We can do that . Input: Das ist zu knapp , weil ich ab dem dritten in Kaiserslautern bin . Genaugenommen nur am dritten . Wie ware es denn am ahm Samstag , dem zehnten Februar ? MonS: That is too tight , because I from the third in Kaiserslautern . In fact only on the third . How about ahm Saturday , the tenth of February ? QmS: That is too tight , because I am from the third in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? IbmS: That is too tight , from the third because I will be in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? Input: Wenn Sie dann noch den siebzehnten konnten , ware das toll , ja . MonS: If you then also the seventeenth could , would be the great , yes . QmS: If you could then also the seventeenth , that would be great , yes . IbmS: Then if you could even take seventeenth , that would be great , yes . Input: Ja , das kommt mir sehr gelegen . Machen wir es dann am besten so . MonS: Yes , that suits me perfectly . Do we should best like that . QmS: Yes , that suits me fine . We do it like that then best . IbmS: Yes , that suits me fine . We should best do it like that .","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.5 Connectives with scopes.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Overall, language modeling significantly impacts decoder performance.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We again assume a training set of n examples {x1 . xri} where the first m examples have labels {y1 ... yin}, and the last (n \u2014 m) examples are unlabeled.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"12 For English, our Evalb implementation is identical to the most recent reference (EVALB20080701).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"All the links in the \u00e2\u0080\u009cCC-domain are shown in Step 4 in subsection 3.2.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"While we cannot prove there are no such useful features on which one should condition trust, we can give some insight into why the features we explored offered no gain.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"We are given a source string fJ 1 = f1:::fj :::fJ of length J, which is to be translated into a target string eI 1 = e1:::ei:::eI of length I. Among all possible target strings, we will choose the string with the highest probability: ^eI 1 = arg max eI 1 fPr(eI 1jfJ 1 )g = arg max eI 1 fPr(eI 1) Pr(fJ 1 jeI 1)g : (1) The argmax operation denotes the search problem, i.e. the generation of the output sentence in the target language.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Given an anaphor, BABAR identifies the caseframe that would extract it from its sentence.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Altun et al. (2005) proposed a technique that uses graph based similarity between labeled and unlabeled parts of structured data in a discriminative framework for semi-supervised learning.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"This distributional sparsity of syntactic tags is not unique to English 1 The source code for the work presented in this paper is available at http:\/\/groups.csail.mit.edu\/rbg\/code\/typetagging\/.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We use label propagation in two stages to generate soft labels on all the vertices in the graph.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Denote the unthresholded classifiers after t \u2014 1 rounds by git\u20141 and assume that it is the turn for the first classifier to be updated while the second one is kept fixed.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Benchmarks use the package\u2019s binary format; our code is also the fastest at building a binary file.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Dynamic programming efficiently scores many hypotheses by exploiting the fact that an N-gram language model conditions on at most N \u2212 1 preceding words.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We create equivalence classes for verb, noun, and adjective POS categories.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Figure 1 reveals that an event that \u00e2\u0080\u009cdamaged\u00e2\u0080\u009d objects may also cause injuries; a disaster that \u00e2\u0080\u009coccurred\u00e2\u0080\u009d may be investigated to find its \u00e2\u0080\u009ccause\u00e2\u0080\u009d; a disaster may \u00e2\u0080\u009cwreak\u00e2\u0080\u009d havoc as it \u00e2\u0080\u009ccrosses\u00e2\u0080\u009d geographic regions; and vehicles that have a \u00e2\u0080\u009cdriver\u00e2\u0080\u009d may also \u00e2\u0080\u009ccarry\u00e2\u0080\u009d items.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This is not to say that a set of standards by which a particular segmentation would count as correct and another incorrect could not be devised; indeed, such standards have been proposed and include the published PRCNSC (1994) and ROCLING (1993), as well as the unpublished Linguistic Data Consortium standards (ca.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Then each arc of D maps either from an element of H to an element of p, or from E-i.e., the empty string-to an element of P. More specifically, each word is represented in the dictionary as a sequence of arcs, starting from the initial state of D and labeled with an element 5 of Hxp, which is terminated with a weighted arc labeled with an element of Ex P. The weight represents the estimated cost (negative log probability) of the word.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We use two different similarity functions to define the edge weights among the foreign vertices and between vertices from different languages.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Like in the co-reference annotation, G\u00c2\u00a8otze\u00e2\u0080\u0099s proposal has been applied by two annotators to the core corpus but it has not been systematically evaluated yet.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"73 81.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We use a simple TF\/IDF method to measure the topicality of words.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We used these three parsers to explore parser combination techniques.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Projectivizing a dependency graph by lifting nonprojective arcs is a nondeterministic operation in the general case.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This is because our corpus is not annotated, and hence does not distinguish between the various words represented by homographs, such as, which could be \/adv jiangl 'be about to' orInc jiang4 '(military) general'-as in 1j\\xiao3jiang4 'little general.'","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Our work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Our System Wang, Li, and Chang a. 1\\!f!IP Eflltii \/1\\!f!J:P $1til I b. agm: I a m: c. 5 Bf is Bf 1 d. \"*:t: w _t ff 1 \"* :t: w_tff 1 g., , Transliteration\/Translation chen2zhongl-shenl qu3 'music by Chen Zhongshen ' huang2rong2 youlyoul de dao4 'Huang Rong said soberly' zhangl qun2 Zhang Qun xian4zhang3 you2qingl shang4ren2 hou4 'after the county president You Qing had assumed the position' lin2 quan2 'Lin Quan' wang2jian4 'Wang Jian' oulyang2-ke4 'Ouyang Ke' yinl qi2 bu4 ke2neng2 rong2xu3 tai2du2 er2 'because it cannot permit Taiwan Independence so' silfa3-yuan4zhang3 lin2yang2-gang3 'president of the Judicial Yuan, Lin Yanggang' lin2zhangl-hu2 jiangl zuo4 xian4chang3 jie3shuol 'Lin Zhanghu will give an ex\u00c2\u00ad planation live' jin4\/iang3 nian2 nei4 sa3 xia4 de jinlqian2 hui4 ting2zhi3 'in two years the distributed money will stop' gaoltangl da4chi2 ye1zi0 fen3 'chicken stock, a tablespoon of coconut flakes' you2qingl ru4zhu3 xian4fu3 lwu4 'after You Qing headed the county government' Table 5 Performance on morphological analysis.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"First we mark any node that dominates (at any level) a verb sider POS tags when pre-terminals are the only intervening nodes between the nucleus and its bracketing (e.g., unaries, base NPs).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"A straightforward way to find the shortest tour is by trying all possible permutations of the n cities.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The rest of the paper is structured as follows.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This has the potential drawback of increasing the number of features, which can make MERT less stable (Foster and Kuhn, 2009).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Our monolingual similarity function (for connecting pairs of foreign trigram types) is the same as the one used by Subramanya et al. (2010).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The linear and nonerasing assumptions about the operations discussed in Section 4.1 require that each z, and yk is used exactly once to define the strings zi, ,z1,3.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"And then there are decisions that systems typically hard-wire, because the linguistic motivation for making them is not well understood yet.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Statistical methods seem particularly applicable to the problem of unknown-word identification, especially for constructions like names, where the linguistic constraints are minimal, and where one therefore wants to know not only that a particular se\u00c2\u00ad quence of hanzi might be a name, but that it is likely to be a name with some probabil\u00c2\u00ad ity.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The sentences in the corpus were tagged by a transformation-based chunker and an NE tagger.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"1 74.5 56.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Intuitively, it places more weight on OUT when less evidence from IN is available.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Once we figure out the important word (e.g. keyword), we believe we can capture the meaning of the phrase by the keyword.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"On each step CoBoost searches for a feature and a weight so as to minimize either 40 or 40.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For each set, the phrases with bracketed frequencies are considered not paraphrases in the set.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The search starts in hypothesis (f;g; 0) and ends in the hypotheses (f1; ; Jg; j), with j 2 f1; ; Jg.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"While Berg-Kirkpatrick et al.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We extend the Matsoukas et al approach in several ways.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Using the terminology of Kahane et al. (1998), we say that jedna is the syntactic head of Z, while je is its linear head in the projectivized representation.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"These tag distributions are used to initialize the label distributions over the English vertices in the graph.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Subsets of partial hypotheses with coverage sets C of increasing cardinality c are processed.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Finally, we incorporate the instance-weighting model into a general linear combination, and learn weights and mixing parameters simultaneously. where c\u03bb(s, t) is a modified count for pair (s, t) in OUT, u(s|t) is a prior distribution, and y is a prior weight.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"With the additional assumptions, inspired by Rounds (1985), we can show that members of this class can be recognized in polynomial time.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"This is especially true in the case of quotations\u00e2\u0080\u0094which are common in the ATB\u00e2\u0080\u0094where (1) will follow a verb like (2) (Figure 1).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Our monolingual similarity function (for connecting pairs of foreign trigram types) is the same as the one used by Subramanya et al. (2010).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"This aspect of the formalism is both linguistically and computationally important.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The dictionary sizes reported in the literature range from 17,000 to 125,000 entries, and it seems reasonable to assume that the coverage of the base dictionary constitutes a major factor in the performance of the various approaches, possibly more important than the particular set of methods used in the segmentation.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The resulting algorithm is depicted in Table 1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Methods that allow multiple segmentations must provide criteria for choosing the best segmentation.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Many researchers have developed coreference resolvers, so we will only discuss the methods that are most closely related to BABAR.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Better grammars are shown here to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","label":1},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Inspection of the data shows that at n = 2500, the two classifiers both give labels on 44,281 (49.2%) of the unlabeled examples, and give the same label on 99.25% of these cases.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The Prague Dependency Treebank (PDT) consists of more than 1M words of newspaper text, annotated on three levels, the morphological, analytical and tectogrammatical levels (Haji\u02c7c, 1998).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The set is then compared with the set generated from the Penn Treebank parse to determine the precision and recall.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We have modified Moses (Koehn et al., 2007) to keep our state with hypotheses; to conserve memory, phrases do not keep state.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In that work, mutual information was used to decide whether to group adjacent hanzi into two-hanzi words.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Month Names, Days of the Week, and Numbers: If is initCaps and is one of January, February, . . .","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(Riloff and Shepherd 97) describe a bootstrapping approach for acquiring nouns in particular categories (such as &quot;vehicle&quot; or &quot;weapon&quot; categories).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Case and Zone: If the token starts with a capital letter (initCaps), then an additional feature (init- Caps, zone) is set to 1.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"This is because in many cases, the use of hyphens can be considered to be optional (e.g., third-quarter or third quarter).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The Potsdam Commentary Corpus","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"All our results are obtained by using only the official training data provided by the MUC conferences.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Independence of paths at this level reflects context freeness of rewriting and suggests why they can be recognized efficiently.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Other approaches encode sparsity as a soft constraint.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A moment's reflection will reveal that things are not quite that simple.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"As suggested at the end of Section 3, the restrictions that have been specified in the definition of LCFRS's suggest that they can be efficiently recognized.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This WFST represents the segmentation of the text into the words AB and CD, word boundaries being marked by arcs mapping between f and part-of-speech labels.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The segmenter will give both analyses :1 cai2 neng2 'just be able,' and ?]cai2neng2 'talent,' but the latter analysis is preferred since splitting these two morphemes is generally more costly than grouping them.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Using less training data than other systems, our NER is able to perform as well as other state-of-the-art NERs.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The reason why we did not train with both MUC6 and MUC7 training data at the same time is because the task specifications for the two tasks are not identical.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"However, 1 http:\/\/maxent.sourceforge.net 3.2 Testing.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"This is summarized in Equation 5.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Hence, s < k. But by addition of the votes on the two parses, s > 2N-11> k, a contradiction.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"And indeed, converging on annotation guidelines is even more difficult than it is with co-reference.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"2.2.1 The Caseframe Representation Information extraction (IE) systems use extraction patterns to identify noun phrases that play a specific role in 1 Our implementation only resolves NPs that occur in the same document, but in retrospect, one could probably resolve instances of the same existential NP in different documents too.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"At present, the \u00e2\u0080\u0098Potsdam Commentary Corpus\u00e2\u0080\u0099 (henceforth \u00e2\u0080\u0098PCC\u00e2\u0080\u0099 for short) consists of 170 commentaries from Ma\u00c2\u00a8rkische Allgemeine Zeitung, a German regional daily.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"JI!","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In Section 6 we dis\u00c2\u00ad cuss other issues relating to how higher-order language models could be incorporated into the model.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This decreases the statistical significance of our results compared to those studies.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In speech recognition the arcs of the lattice are typically weighted in order to indicate the probability of specific transitions.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"There is no global pruning.","target_text":"The type of alignment we have considered so far requires the same length for source and target sentence, i.e. I = J. Evidently, this is an unrealistic assumption, therefore we extend the concept of inverted alignments as follows: When adding a new position to the coverage set C, we might generate either \u00c3\u0086 = 0 or \u00c3\u0086 = 1 new target words.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"None of the parsers produce parses with crossing brackets, so none of them votes for both of the assumed constituents.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Ex: Mr. Cristiani is the president ...","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Mandarin exhibits several such processes, including A-not-A question formation, il\u00c2\u00ad lustrated in (3a), and adverbial reduplication, illustrated in (3b): 3.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"This overview is illustrated in Figure 1.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"In all of our experiments, the binary file (whether mapped or, in the case of most other packages, interpreted) is loaded into the disk cache in advance so that lazy mapping will never fault to disk.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The resulting algorithm is depicted in Table 1.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Finally, the concatenated 5 * 20% output is used to train the reference resolution component.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"They are set to fixed constants.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"They demonstrated this with the comparison of statistical systems against (a) manually post-edited MT output, and (b) a rule-based commercial system.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"All the links in the \u00e2\u0080\u009cCC-domain are shown in Step 4 in subsection 3.2.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Another attempt at using global information can be found in (Borthwick, 1999).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We evaluate the time and memory consumption of each data structure by computing perplexity on 4 billion tokens from the English Gigaword corpus (Parker et al., 2009).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We have provided methods for handling certain classes of unknown words, and models for other classes could be provided, as we have noted.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Searching a probing hash table consists of hashing the key, indexing the corresponding bucket, and scanning buckets until a matching key is found or an empty bucket is encountered, in which case the key does not exist in the table.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Saving state allows our code to walk the data structure exactly once per query.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"(2010) consistently outperforms ours on English, we obtain substantial gains across other languages.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Examples will usually be accompanied by a translation, plus a morpheme-by-morpheme gloss given in parentheses whenever the translation does not adequately serve this purpose.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"During development, we sensed that the Recency and Syn- role KSs did not deserve to be on equal footing with the other KSs because their knowledge was so general.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"yu2 'fish.'","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"constitute names, since we have only their segmentation, not the actual classification of the segmented words.","label":0},{"source_text":"There is no global pruning.","target_text":"When aligning the words in parallel texts (for language pairs like SpanishEnglish, French-English, ItalianGerman,...), we typically observe a strong localization effect.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Two sets of examples from Gan are given in (1) and (2) (:::::: Gan's Appendix B, exx.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The prediction based on these features is a knearest neighbor classification, using the IB1 algorithm and k = 5, the modified value difference metric (MVDM) and class voting with inverse distance weighting, as implemented in the TiMBL software package (Daelemans et al., 2003).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Lemma: If the number of votes required by constituent voting is greater than half of the parsers under consideration the resulting structure has no crossing constituents.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"At first glance, this seems only peripherally related to our work, since the specific\/general distinction is made for features rather than instances.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"These systems rely on a training corpus that has been manually annotated with coreference links.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Instead of offsetting new topics with punctuation, writers of MSA in sert connectives such as \u00ef\u00bf\u00bd wa and \u00ef\u00bf\u00bd fa to link new elements to both preceding clauses and the text as a whole.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The problem of &quot;noise&quot; items that do not fall into any of the three categories also needs to be addressed.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We used these three parsers to explore parser combination techniques.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"For example, in the phrase \u00e2\u0080\u009cCompany-A last week purchased rival Marshalls from Company-B\u00e2\u0080\u009d, the purchased company is Marshalls, not Company-B.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In considering this aspect of a formalism, we hope to better understand the relationship between the structural descriptions generated by the grammars of a formalism, and the properties of semilinearity and polynomial recognizability.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We use a simple TF\/IDF method to measure the topicality of words.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"It was filtered to retain the top 30 translations for each source phrase using the TM part of the current log-linear model.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Further, it needs extra pointers in the trie, increasing model size by 40%.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"For each trigram type x2 x3 x4 in a sequence x1 x2 x3 x4 x5, we count how many times that trigram type co-occurs with the different instantiations of each concept, and compute the point-wise mutual information (PMI) between the two.5 The similarity between two trigram types is given by summing over the PMI values over feature instantiations that they have in common.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example: McCann initiated a new global system.","label":0},{"source_text":"Here we present two algorithms.","target_text":"There has been additional recent work on inducing lexicons or other knowledge sources from large corpora.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Our System Wang, Li, and Chang a. 1\\!f!IP Eflltii \/1\\!f!J:P $1til I b. agm: I a m: c. 5 Bf is Bf 1 d. \"*:t: w _t ff 1 \"* :t: w_tff 1 g., , Transliteration\/Translation chen2zhongl-shenl qu3 'music by Chen Zhongshen ' huang2rong2 youlyoul de dao4 'Huang Rong said soberly' zhangl qun2 Zhang Qun xian4zhang3 you2qingl shang4ren2 hou4 'after the county president You Qing had assumed the position' lin2 quan2 'Lin Quan' wang2jian4 'Wang Jian' oulyang2-ke4 'Ouyang Ke' yinl qi2 bu4 ke2neng2 rong2xu3 tai2du2 er2 'because it cannot permit Taiwan Independence so' silfa3-yuan4zhang3 lin2yang2-gang3 'president of the Judicial Yuan, Lin Yanggang' lin2zhangl-hu2 jiangl zuo4 xian4chang3 jie3shuol 'Lin Zhanghu will give an ex\u00c2\u00ad planation live' jin4\/iang3 nian2 nei4 sa3 xia4 de jinlqian2 hui4 ting2zhi3 'in two years the distributed money will stop' gaoltangl da4chi2 ye1zi0 fen3 'chicken stock, a tablespoon of coconut flakes' you2qingl ru4zhu3 xian4fu3 lwu4 'after You Qing headed the county government' Table 5 Performance on morphological analysis.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"6 Results and Analysis.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"For English POS tagging, BergKirkpatrick et al. (2010) found that this direct gradient method performed better (>7% absolute accuracy) than using a feature-enhanced modification of the Expectation-Maximization (EM) algorithm (Dempster et al., 1977).8 Moreover, this route of optimization outperformed a vanilla HMM trained with EM by 12%.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Keywords with more than one word In the evaluation, we explained that \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d are considered paraphrases.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Recently, this topic has been getting more attention, as is evident from the Paraphrase Workshops in 2003 and 2004, driven by the needs of various NLP applications.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Global features are extracted from other occurrences of the same token in the whole document.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Both parametric and non-parametric models are explored.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The results in Table 2 were achieved on the development set.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"However, in practice, unknown word models also make the distribution improper.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"In our grammar, features are realized as annotations to basic category labels.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"It then computes a normalized Levenshtein edit distance between the extracted chain and the reference.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Alon Lavie advised on this work.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Since guess and gold trees may now have different yields, the question of evaluation is complex.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The results are given in Table 4.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Acknowledgments We thank Steven Bethard, Evan Rosen, and Karen Shiells for material contributions to this work.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"'Malaysia.'","label":0},{"source_text":"There is no global pruning.","target_text":"A position is presented by the word at that position.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Motivated by these questions, we significantly raise baselines for three existing parsing models through better grammar engineering.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The 4th block contains instance-weighting models trained on all features, used within a MAP TM combination, and with a linear LM mixture.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"4.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"2.1 Overview.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This suggests that the backoff model is as reasonable a model as we can use in the absence of further information about the expected cost of a plural form.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"One of such approaches uses comparable documents, which are sets of documents whose content are found\/known to be almost the same, such as different newspaper stories about the same event [Shinyama and Sekine 03] or different translations of the same story [Barzilay 01].","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Let s = a + b.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Indeed there are several open issues.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Zt can be written as follows Following the derivation of Schapire and Singer, providing that W+ > W_, Equ.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This is a standard adaptation problem for SMT.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The samples from each corpus were independently evaluated.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"\u00ce\u00b2 is the shared hyperparameter for the tag assignment prior and word feature multinomials.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, suppose the current model assigns a belief value of .60 to {A, B}, meaning that it is 60% sure that the correct hypothesis is either A or B. Then new evidence arrives with a belief value of .70 assigned 5 Initially there are no competing hypotheses because all hypotheses are included in \u00ce\u00b8 by definition.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"First, we will describe their method and compare it with our method.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Maamouri et al.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Hyperparameters Our model has two Dirichlet concentration hyperparameters: \u00ce\u00b1 is the shared hyperparameter for the token-level HMM emission and transition distributions.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The table makes clear that enriching our grammar improves the syntactic performance as well as morphological disambiguation (segmentation and POS tagging) accuracy.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This best instance-weighting model beats the equivalant model without instance weights by between 0.6 BLEU and 1.8 BLEU, and beats the log-linear baseline by a large margin.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In the denomi 11 We have two such lists, one containing about 17,000 full names, and another containing frequencies of.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"11 taTweel (-) is an elongation character used in Arabic script to justify text.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Finally, model GTv = 2 includes parent annotation on top of the various state-splits, as is done also in (Tsarfaty and Sima\u2019an, 2007; Cohen and Smith, 2007).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In this paper, we have presented a new, e\u00c3\u0086cient DP-based search procedure for statistical machine translation.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"19 We note that it is not always clear in Wang, Li, and Chang's examples which segmented words.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Word Re-ordering and DP-based Search in Statistical Machine Translation","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Zt can be written as follows Following the derivation of Schapire and Singer, providing that W+ > W_, Equ.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"\u00e2\u0080\u009cThe gun\u00e2\u0080\u009d will be extracted by the caseframe \u00e2\u0080\u009cfired <patient>\u00e2\u0080\u009d.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"What ought to be developed now is an annotation tool that can make use of the format, allow for underspecified annotations and visualize them accordingly.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This best instance-weighting model beats the equivalant model without instance weights by between 0.6 BLEU and 1.8 BLEU, and beats the log-linear baseline by a large margin.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We used the MUC4 terrorism corpus (MUC4 Proceedings, 1992) and news articles from the Reuter\u00e2\u0080\u0099s text collection8 that had a subject code corresponding to natural disasters.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Different annotations of the same text are mapped into the same data structure, so that search queries can be formulated across annotation levels.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Unsupervised learning approaches appear to be a natural solution to this problem, as they require only unannotated text for training models.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Since these are distinct syntactic units, they are typically segmented.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"2 Chinese ?l* han4zi4 'Chinese character'; this is the same word as Japanese kanji..","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"able at http:\/\/nlp.stanford.edu\/projects\/arabic.shtml.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We compare the manually annotated grammar, which we incorporate into the Stanford parser, to both the Berkeley (Petrov et al., 2006) and Bikel (Bikel, 2004) parsers.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In this subsection, we will report the results of the experiment, in terms of the number of words, phrases or clusters.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Much could be done to further reduce memory consumption.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The word joining is done on the basis of a likelihood criterion.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Note that Chang, Chen, and Chen (1991), in addition to word-frequency information, include a constraint-satisfication model, so their method is really a hybrid approach.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"This modification brings the method closer to the DL-CoTrain algorithm described earlier, and is motivated by the intuition that all three labels should be kept healthily populated in the unlabeled examples, preventing one label from dominating \u2014 this deserves more theoretical investigation.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"raphy: A ren2 'person' is a fairly uncontroversial case of a monographemic word, and rplil zhong1guo2 (middle country) 'China' a fairly uncontroversial case of a di\u00c2\u00ad graphernic word.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"AdaBoost.MH can be applied to the problem using these pseudolabels in place of supervised examples.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"genitiveMark indicates recursive NPs with a indefinite nominal left daughter and an NP right daughter.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"73 81.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The state function is integrated into the query process so that, in lieu of the query p(wnjwn\u22121 1 ), the application issues query p(wnjs(wn\u22121 1 )) which also returns s(wn1 ).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"(2010) reports the best unsupervised results for English.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For humans, this characteristic can impede the acquisition of literacy.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The transition from f, to a final state transduces c to the grammatical tag PL with cost cost(unseen(f,)): cost(i\u00c2\u00a5JJ1l.ir,) == cost(i\u00c2\u00a5JJ1l.)","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We believe that the underlying principles of the maximum entropy framework are suitable for exploiting information from diverse sources.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The hash variant is a reverse trie with hash tables, a more memory-efficient version of SRILM\u2019s default.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We compared the ATB5 to tree- banks for Chinese (CTB6), German (Negra), and English (WSJ) (Table 4).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Our results outperform strong unsupervised baselines as well as approaches that rely on direct projections, and bridge the gap between purely supervised and unsupervised POS tagging models.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"To do this, the x's and y's are stored in the next 2ni + 2n2 tapes, and M goes to a universal state.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Unsupervised Models for Named Entity Classification Collins","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"3 60.7 50.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The Bikel GoldPOS configuration only supplies the gold POS tags; it does not force the parser to use them.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"If a phrase does not contain any keywords, the phrase is discarded.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"segmentation (Table 2).","label":0},{"source_text":"Here we present two algorithms.","target_text":"Again, this deserves further investigation.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"There is a sizable literature on Chinese word segmentation: recent reviews include Wang, Su, and Mo (1990) and Wu and Tseng (1993).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"An initial step of any text\u00c2\u00ad analysis task is the tokenization of the input into words.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Ex: Mr. Cristiani is the president ...","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"( b ) s u p p o r t s c a n d i d a t e i f s e l e c t e d s e m a n t i c t a g s m a t c h t h o s e o f t h e a n a p h o r . Le xic al computes degree of lexical overlap b e t w e e n t h e c a n d i d a t e a n d t h e a n a p h o r . Re cen cy computes the relative distance between the c a n d i d a t e a n d t h e a n a p h o r . Sy nR ole computes relative frequency with which the c a n d i d a t e \u00e2\u0080\u0099 s s y n t a c t i c r o l e o c c u r s i n r e s o l u t i o n s . Figure 4: General Knowledge Sources The Lexical KS returns 1 if the candidate and anaphor are identical, 0.5 if their head nouns match, and 0 otherwise.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The size of TRIE is particularly sensitive to F1092 c11, so vocabulary filtering is quite effective at reducing model size.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Such constraints are derived from training data, expressing some relationship between features and outcome.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The morphological anal\u00c2\u00adysis itself can be handled using well-known techniques from finite-state morphol 9 The initial estimates are derived from the frequencies in the corpus of the strings of hanzi making up.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"This is the parse that is closest to the centroid of the observed parses under the similarity metric.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"There are two differences between this method and the DL-CoTrain algorithm: spelling and contextual features, alternating between labeling and learning with the two types of features.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This is an issue that we have not addressed at the current stage of our research.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We allow any number of bits from 2 to 25, unlike IRSTLM (8 bits) and BerkeleyLM (17\u221220 bits).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We respond to this on the one hand with a format for its underspecification (see 2.4) and on the other hand with an additional level of annotation that attends only to connectives and their scopes (see 2.5), which is intended as an intermediate step on the long road towards a systematic and objective treatment of rhetorical structure.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The likelihood of the observed data under the model is where P(yi, xi) is defined as in (9).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Any opinions, findings, conclusions, or recommendations expressed in this paper are those of the authors, and do not necessarily reflect the views of the funding organizations.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"0 55.3 34.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"(c) Coordination ambiguity is shown in dependency scores by e.g., \u00e2\u0088\u0097SSS R) and \u00e2\u0088\u0097NP NP NP R).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"This withdrawal by the treasury secretary is understandable, though.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"com t 600 Mountain Avenue, 2c278, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The ratio of buckets to entries is controlled by space multiplier m > 1.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The features are used to represent each example for the learning algorithm.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Next, we represent the input sentence as an unweighted finite-state acceptor (FSA) I over H. Let us assume the existence of a function Id, which takes as input an FSA A, and produces as output a transducer that maps all and only the strings of symbols accepted by A to themselves (Kaplan and Kay 1994).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We refer to this process as Reliable Case Resolution because it involves finding cases of anaphora that can be easily resolved with their antecedents.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(e.g., N.Y. would contribute this feature, IBM would not). nonalpha=x Appears if the spelling contains any characters other than upper or lower case letters.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"This withdrawal by the treasury secretary is understandable, though.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"By applying an inverse transformation to the output of the parser, arcs with non-standard labels can be lowered to their proper place in the dependency graph, giving rise 1The dependency graph has been modified to make the final period a dependent of the main verb instead of being a dependent of a special root node for the sentence. to non-projective structures.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This allow the learners to &quot;bootstrap&quot; each other by filling the labels of the instances on which the other side has abstained so far.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In these examples, the names identified by the two systems (if any) are underlined; the sentence with the correct segmentation is boxed.19 The differences in performance between the two systems relate directly to three issues, which can be seen as differences in the tuning of the models, rather than repre\u00c2\u00ad senting differences in the capabilities of the model per se.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"genitiveMark indicates recursive NPs with a indefinite nominal left daughter and an NP right daughter.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"A corpus of German newspaper commentaries has been assembled at Potsdam University, and annotated with different linguistic information, to different degrees.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In many cases, inspection of either the spelling or context alone is sufficient to classify an example.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Co-occurrences among the particles themselves are subject to further syntactic and lexical constraints relative to the stem.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Finally, as\u00c2\u00ad suming a simple bigram backoff model, we can derive the probability estimate for the particular unseen word i\u00c2\u00a51J1l.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Each name class is subdivided into 4 sub-classes, i.e., N begin, N continue, N end, and N unique.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"mein 5.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Purely statistical approaches have not been very popular, and so far as we are aware earlier work by Sproat and Shih (1990) is the only published instance of such an approach.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We adopted this state-of-the-art model because it makes it easy to experiment with various ways of incorporating our novel constraint feature into the log-linear emission model.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In practice, this sparsity constraint is difficult to incorporate in a traditional POS induction system (Me\u00c2\u00b4rialdo, 1994; Johnson, 2007; Gao and Johnson, 2008; Grac\u00c2\u00b8a et al., 2009; Berg-Kirkpatrick et al., 2010).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"2 for the accuracy of the different methods.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Figure 3 shows examples of semantic expectations that were learned.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Section 2 describes our baseline techniques for SMT adaptation, and section 3 describes the instance-weighting approach.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Keys to the table are hashed, using for example Austin Appleby\u2019s MurmurHash2, to integers evenly distributed over a large range.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"No: Predecessor coverage set Successor coverage set 1 (f1; ;mg n flg ; l0) !","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The features are used to represent each example for the learning algorithm.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Our first algorithm is similar to Yarowsky's, but with some important modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"While it is possible to derive a closed form solution for this convex objective function, it would require the inversion of a matrix of order |Vf|.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The rationale for treating these semantic labels differently is that they are specific and reliable (as opposed to the WordNet classes, which are more coarse and more noisy due to polysemy).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"A Single Generative Model for Joint Morphological Segmentation and Syntactic Parsing","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"A dynamic programming recursion similar to the one in Eq. 2 is evaluated.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Previous work has attempted to incorporate such constraints into token-level models via heavy-handed modifications to inference procedure and objective function (e.g., posterior regularization and ILP decoding) (Grac\u00c2\u00b8a et al., 2009; Ravi and Knight, 2009).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Instead, we focused on phrases and set the frequency threshold to 2, and so were able to utilize a lot of phrases while minimizing noise.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"To bridge this gap, we consider a practically motivated scenario, in which we want to leverage existing resources from a resource-rich language (like English) when building tools for resource-poor foreign languages.1 We assume that absolutely no labeled training data is available for the foreign language of interest, but that we have access to parallel data with a resource-rich language.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In this paper, Section 2 begins by explaining how contextual role knowledge is represented and learned.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The effect of the pruning threshold t0 is shown in Table 5.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The 14 general-language features embody straightforward cues: frequency, \u201ccentrality\u201d as reflected in model scores, and lack of burstiness.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Words and punctuation that appear in brackets are considered optional.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"3.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"This causes a problem for reverse trie implementations, including SRILM itself, because it leaves n+1-grams without an n-gram node pointing to them.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"0 D\/ nc 5.0 The minimal dictionary encoding this information is represented by the WFST in Figure 2(a).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Lexical-knowledge-based approaches that include statistical information generally presume that one starts with all possible segmentations of a sentence, and picks the best segmentation from the set of possible segmentations using a probabilistic or cost\u00c2\u00ad based scoring mechanism.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We used a standard one-pass phrase-based system (Koehn et al., 2003), with the following features: relative-frequency TM probabilities in both directions; a 4-gram LM with Kneser-Ney smoothing; word-displacement distortion model; and word count.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"This PCFG is incorporated into the Stanford Parser, a factored model that chooses a 1-best parse from the product of constituency and dependency parses.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"May 1995).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"971,746 sentences of New York Times text were parsed using the parser of (Collins 96).1 Word sequences that met the following criteria were then extracted as named entity examples: whose head is a singular noun (tagged NN).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"This PCFG is incorporated into the Stanford Parser, a factored model that chooses a 1-best parse from the product of constituency and dependency parses.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The paper explains the design decisions taken in the annotations, and describes a number of applications using this corpus with its multi-layer annotation.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"A token that is allCaps will also be initCaps.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A related point is that mutual information is helpful in augmenting existing electronic dictionaries, (cf.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Acronyms (ACRO): Words made up of all capitalized letters in the text zone will be stored as acronyms (e.g., IBM).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Future extensions of the system might include: 1) An extended translation model, where we use more context to predict a source word.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"25 16.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Our annotators pointed out that very often they made almost random decisions as to what relation to choose, and where to locate the boundary of a span.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The inventions disclosed herein are the subject of a patent application owned by the University of Utah and licensed on an exclusive basis to Attensity Corporation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The method just described segments dictionary words, but as noted in Section 1, there are several classes of words that should be handled that are not found in a standard dictionary.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"As far as we know, no other NERs have used information from the whole document (global) as well as information within the same sentence (local) in one framework.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Given a document to process, BABAR uses four modules to perform coreference resolution.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We define a symmetric similarity function K(uZ7 uj) over two foreign language vertices uZ7 uj E Vf based on the co-occurrence statistics of the nine feature concepts given in Table 1.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In (b) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the kidnapping victims, but in (c) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the armed men.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"A list of words occurring more than 10 times in the training data is also collected (commonWords).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We have shown that the maximum entropy framework is able to use global information directly.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"For robustness reasons, the parser may output a set of dependency trees instead of a single tree. most dependent of the next input token, dependency type features are limited to tokens on the stack.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We then use linguistic and annotation insights to develop a manually annotated grammar for Arabic (\u00c2\u00a74).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"0 D\/ nc 5.0 The minimal dictionary encoding this information is represented by the WFST in Figure 2(a).","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The goal of machine translation is the translation of a text given in some source language into a target language.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"One may argue with these efforts on normalization, and ultimately their value should be assessed by assessing their impact on inter-annotator agreement.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"1 53.8 47.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Thus, if one wants to segment words-for any purpose-from Chinese sentences, one faces a more difficult task than one does in English since one cannot use spacing as a guide.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"(7) is at 0 when: 1) Vi : sign(gi (xi)) = sign(g2 (xi)); 2) Ig3(xi)l oo; and 3) sign(gi (xi)) = yi for i = 1, , m. In fact, Zco provides a bound on the sum of the classification error of the labeled examples and the number of disagreements between the two classifiers on the unlabeled examples.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Statistical NERs usually find the sequence of tags that maximizes the probability , where is the sequence of words in a sentence, and is the sequence of named-entity tags assigned to the words in . Attempts have been made to use global information (e.g., the same named entity occurring in different sentences of the same document), but they usually consist of incorporating an additional classifier, which tries to correct the errors in the output of a first NER (Mikheev et al., 1998; Borthwick, 1999).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"For each feature type f and tag t, a multinomial \u00cf\u0088tf is drawn from a symmetric Dirichlet distribution with concentration parameter \u00ce\u00b2.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In MUC6 and MUC7, the named entity task is defined as finding the following classes of names: person, organization, location, date, time, money, and percent (Chinchor, 1998; Sundheim, 1995) Machine learning systems in MUC6 and MUC 7 achieved accuracy comparable to rule-based systems on the named entity task.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Conditioned on T , features of word types W are drawn.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Text generation, or at least the two phases of text planning and sentence planning, is a process driven partly by well-motivated choices (e.g., use this lexeme X rather than that more colloquial near-synonym Y ) and partly by con tation like that of PCC can be exploited to look for correlations in particular between syntactic structure, choice of referring expressions, and sentence-internal information structure.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Therefore, performance is more closely tied to the underlying data structure than to the cache.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"0 57.3 51.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"com t 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"2 61.7 64.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This implies, therefore, that a major factor in the performance of a Chinese segmenter is the quality of the base dictionary, and this is probably a more important factor-from the point of view of performance alone-than the particular computational methods used.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Extract NE instance pairs with contexts First, we extract NE pair instances with their context from the corpus.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"3 58.3 40.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For pronouns, however, all of the knowledge sources increased recall, often substantially, and with little if any decrease in precision.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Other packages walk their respective data structures once to find wnf and again to find {b(wn\u22121 i )}f\u22121 i=1if necessary.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We thank members of the MIT NLP group for their suggestions and comments.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"gaolbu4-gaolxing4 (hap-not-happy) 'happy?'","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Moreover, the Stanford parser achieves the most exact Leaf Ancestor matches and tagging accuracy that is only 0.1% below the Bikel model, which uses pre-tagged input.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Storing state therefore becomes a time-space tradeoff; for example, we store state with partial hypotheses in Moses but not with each phrase.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold, 30.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"However, the accuracy is considerably higher than previously reported results for robust non-projective parsing of Czech, with a best performance of 73% UAS (Holan, 2004).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We evaluate our approach on eight European languages (\u00a76), and show that both our contributions provide consistent and statistically significant improvements.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"A promising direction for future work is to explicitly model a distribution over tags for each word type.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The Bikel GoldPOS configuration only supplies the gold POS tags; it does not force the parser to use them.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This section measures performance on shared tasks in order of increasing complexity: sparse lookups, evaluating perplexity of a large file, and translation with Moses.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We consider the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"1,000 of these were picked at random, and labeled by hand to produce a test set.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"((fii, Q2, Pa) , \u2014\u25a0 (01, i32, 03) , e),(02,e) Oa, en) The path complexity of the tee set generated by a MCTAG is not necessarily context-free.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Previous reports on Chinese segmentation have invariably cited performance either in terms of a single percent-correct score, or else a single precision-recall pair.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Cohen and Smith (2007) later on based a system for joint inference on factored, independent, morphological and syntactic components of which scores are combined to cater for the joint inference task.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"What ought to be developed now is an annotation tool that can make use of the format, allow for underspecified annotations and visualize them accordingly.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"In the label propagation stage, we propagate the automatic English tags to the aligned Italian trigram types, followed by further propagation solely among the Italian vertices. the Italian vertices are connected to an automatically labeled English vertex.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Among these 32 sets, we found the following pairs of sets which have two or more links.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Hence we decided to select ten commentaries to form a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099, for which the entire range of annotation levels was realized, so that experiments with multi-level querying could commence.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"To improve agreement during the revision process, a dual-blind evaluation was performed in which 10% of the data was annotated by independent teams.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"6 Our knowledge sources return some sort of probability estimate, although in some cases this estimate is not especially well-principled (e.g., the Recency KS).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The tokens w are generated by token-level tags t from an HMM parameterized by the lexicon structure.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"As we shall see, most of the linked sets are paraphrases.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"2.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"We assume that the goal in dependency parsing is to construct a labeled dependency graph of the kind depicted in Figure 1.","label":0},{"source_text":"There is no global pruning.","target_text":"13.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Other kinds of productive word classes, such as company names, abbreviations (termed fijsuolxie3 in Mandarin), and place names can easily be 20 Note that 7 in E 7 is normally pronounced as leO, but as part of a resultative it is liao3..","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In section 5, we then evaluate the entire parsing system by training and evaluating on data from the Prague Dependency Treebank.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For each pair we also record the context, i.e. the phrase between the two NEs (Step1).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"This work has been supported as part of the Verbmobil project (contract number 01 IV 601 A) by the German Federal Ministry of Education, Science, Research and Technology and as part of the Eutrans project (ESPRIT project number 30268) by the European Community.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Subsets of partial hypotheses with coverage sets C of increasing cardinality c are processed.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If is unique, then a feature (Unique, Zone) is set to 1, where Zone is the document zone where appears.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For example, the phrase \u00e2\u0080\u009c's New York-based trust unit,\u00e2\u0080\u009d is not a paraphrase of the other phrases in the \u00e2\u0080\u009cunit\u00e2\u0080\u009d set.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"That is, given a choice between segmenting a sequence abc into abc and ab, c, the former will always be picked so long as its cost does not exceed the summed costs of ab and c: while; it is possible for abc to be so costly as to preclude the larger grouping, this will certainly not usually be the case.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For example, the Wang, Li, and Chang system fails on the sequence 1:f:p:]nian2 nei4 sa3 in (k) since 1F nian2 is a possible, but rare, family name, which also happens to be written the same as the very common word meaning 'year.'","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We present a coreference resolver called BABAR that uses contextual role knowledge to evaluate possible antecedents for an anaphor.","label":1},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"(If the TF\/IDF score of that word is below a threshold, the phrase is discarded.)","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Model components cascade, so the row corresponding to +FEATS also includes the PRIOR component (see Section 3).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"99 94.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We use the log-linear tagger of Toutanova et al.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In this case, we have no finite-state restrictions for the search space.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The possible analyses of a surface token pose constraints on the analyses of specific segments.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Can we do . QmS: Yes, wonderful.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The segmenter handles the grouping of hanzi into words and outputs word pronunciations, with default pronunciations for hanzi it cannot group; we focus here primarily on the system's ability to segment text appropriately (rather than on its pronunciation abilities).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"A detailed description of the search procedure used is given in this patent.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We see from these results that the behavior of the parametric techniques are robust in the presence of a poor parser.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Finally, the assignments of PoS tags to OOV segments is subject to language specific constraints relative to the token it was originated from.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This withdrawal by the treasury secretary is understandable, though.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Annotation of syntactic structure for the core corpus has just begun.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We refer to (T , W ) as the lexicon of a language and \u00cf\u0088 for the parameters for their generation; \u00cf\u0088 depends on a single hyperparameter \u00ce\u00b2.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"First, we will describe their method and compare it with our method.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"nan2gual 'pumpkin.'","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"If they knew that the first four words in a hypergraph node would never extend to the left and form a 5-gram, then three or even fewer words could be kept in the backward state.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We extend the Stanford parser to accept pre-generated lattices, where each word is represented as a finite state automaton.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We would like to thank Ryan McDonald for numerous discussions on this topic.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Note that on some examples (around 2% of the test set) CoBoost abstained altogether; in these cases we labeled the test example with the baseline, organization, label.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"0.2 0.1 0.0 -0.1 25 26 27 28 29 30 31 32 -0.2 -0.3 \u2022systran \u2022 ntt 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 20 21 22 23 24 25 26 Fluency Fluency \u2022systran \u2022nrc rali 25 26 27 28 29 30 31 32 0.2 0.1 0.0 -0.1 -0.2 -0.3 cme p \ufffd 20 21 22 23 24 25 26 0.5 0.4 0.3 0.2 0.1 0.0 -0.1 -0.2 -0.3 Figure 14: Correlation between manual and automatic scores for English-French 119 In Domain Out of Domain \u2022upv Adequacy -0.9 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv 23 24 25 26 27 28 29 30 31 32 \u2022upc-mr \u2022utd \u2022upc-jmc \u2022uedin-birch \u2022ntt \u2022rali \u2022uedin-birch 16 17 18 19 20 21 22 23 24 25 26 27 Adequacy \u2022upc-mr 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -1.1 English-Spanish Fluency \u2022ntt \u2022nrc \u2022rali \u2022uedin-birch -0.2 -0.3 -0.5 \u2022upv 16 17 18 19 20 21 22 23 24 25 26 27 -0.4 nr \u2022 rali Fluency -0.4 \u2022upc-mr utd \u2022upc-jmc -0.5 -0.6 \u2022upv 23 24 25 26 27 28 29 30 31 32 0.2 0.1 -0.0 -0.1 -0.2 -0.3 0.3 0.2 0.1 -0.0 -0.1 -0.6 -0.7 Figure 15: Correlation between manual and automatic scores for English-Spanish 120 English-German In Domain Out of Domain Adequacy Adequacy 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 \u2022upv -0.2 -0.3 -0.4 -0.5 -0.6 -0.7 -0.8 -0.9 \u2022upv 0.5 0.4 \u2022systran \u2022upc-mr \u2022 \u2022rali 0.3 \u2022ntt 0.2 0.1 -0.0 -0.1 \u2022systran \u2022upc-mr -0.9 9 10 11 12 13 14 15 16 17 18 19 6 7 8 9 10 11 Fluency 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 \u2022upv -0.5 \u2022upv \u2022systran \u2022upc-mr \u2022 Fluency 0.4 0.3 0.2 0.1 -0.0 -0.1 -0.2 -0.3 -0.4 -0.5 -0.6 \u2022systran \u2022ntt","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The key point is that the second constraint can be remarkably powerful in reducing the complexity of the learning problem.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"(b) POS tagging accuracy is lowest for maSdar verbal nouns (VBG,VN) and adjectives (e.g., JJ).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Whereas Arabic linguistic theory as Saddam (a) Reference (b) Stanford signs (1) and (2) to the class of pseudo verbs 01 +i J>1\u00ef\u00bf\u00bd inna and her sisters since they can beinflected, the ATB conventions treat (2) as a com plementizer, which means that it must be the head of SBAR.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"This paper, however, provides a comprehensive overview of the data collection effort and its current state.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The search starts in the hypothesis (I; f;g; 0).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"and \u00e2\u0080\u009cH\u00e2\u0080\u009d represents \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We proposed an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"There have been other kinds of efforts to discover paraphrase automatically from corpora.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"29 \u00e2\u0080\u0094 95.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We did have a problem classifying some modified noun phrases where the modified phrase does not represent a qualified or restricted form of the head, like \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, as these are both represented by the keyword \u00e2\u0080\u009cchairman\u00e2\u0080\u009d.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In MUC6, the best result is achieved by SRA (Krupka, 1995).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Our analysis begins with a description of syntactic ambiguity in unvocalized MSA text (\u00c2\u00a72).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For the examples given in (1) and (2) this certainly seems possible.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"However, dynamic programming can be used to find the shortest tour in exponential time, namely in O(n22n), using the algorithm by Held and Karp.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The sources of our dictionaries are listed in Table 2.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"One may argue with these efforts on normalization, and ultimately their value should be assessed by assessing their impact on inter-annotator agreement.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We have estimated the performance of IdentiFinder ' 99 at 200K words of training data from the graphs.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We then discuss how we adapt and generalize a boosting algorithm, AdaBoost, to the problem of named entity classification.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Both of the switching techniques, as well as the parametric hybridization technique were also shown to be robust when a poor parser was introduced into the experiments.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We use the HSPELL9 (Har\u2019el and Kenigsberg, 2004) wordlist as a lexeme-based lexicon for pruning segmentations involving invalid segments.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We observe similar trends when using another measure \u00e2\u0080\u0093 type-level accuracy (defined as the fraction of words correctly assigned their majority tag), according to which La ng ua ge M etr ic B K 10 E M B K 10 L B F G S G 10 F EA T S B es t F EA T S M ed ia n E ng lis h 1 1 m 1 4 8 . 3 6 8 . 1 5 6 . 0 7 5 . 5 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 0 . 9 6 6 . 4 4 7 . 8 6 6 . 4 D an is h 1 1 m 1 4 2 . 3 6 6 . 7 4 2 . 6 5 8 . 0 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 2 . 1 6 1 . 2 4 3 . 2 6 0 . 7 D ut ch 1 1 m 1 5 3 . 7 6 7 . 0 5 5 . 1 6 4 . 7 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 6 . 4 6 9 . 0 5 1 . 5 6 7 . 3 Po rtu gu es e 1 1 m 1 5 0 . 8 7 5 . 3 4 3 . 2 7 4 . 8 44 .5 69 .2 6 4 . 1 7 4 . 5 5 6 . 5 7 0 . 1 S pa ni sh 1 1 m 1 \u00e2\u0080\u0093 \u00e2\u0080\u0093 4 0 . 6 7 3 . 2 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 8 . 3 6 8 . 9 5 0 . 0 5 7 . 2 Table 4: Comparison of our method (FEATS) to state-of-the-art methods.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"To evaluate the performance on the segmentation task, we report SEG, the standard harmonic means for segmentation Precision and Recall F1 (as defined in Bar-Haim et al. (2005); Tsarfaty (2006)) as well as the segmentation accuracy SEGTok measure indicating the percentage of input tokens assigned the correct exact segmentation (as reported by Cohen and Smith (2007)).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The following context-free production captures the derivation step of the grammar shown in Figure 7, in which the trees in the auxiliary tree set are adjoined into themselves at the root node (address c).","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We present two algorithms.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We formulate the update as follows: where \u2200ui \u2208 Vf \\ Vfl, \u03b3i(y) and \u03bai are defined as: We ran this procedure for 10 iterations.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We further thank Khalil Simaan (ILLCUvA) for his careful advise concerning the formal details of the proposal.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"We call this pseudoprojective dependency parsing, since it is based on a notion of pseudo-projectivity (Kahane et al., 1998).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"36 79.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Also, \u00e2\u0080\u009cagree\u00e2\u0080\u009d in the CC-domain is not a desirable keyword.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"63 95.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"\u00ce\u00b2 is the shared hyperparameter for the tag assignment prior and word feature multinomials.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Almost all annotators expressed their preference to move to a ranking-based evaluation in the future.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Nonetheless, the results of the comparison with human judges demonstrates that there is mileage being gained by incorporating models of these types of words.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For example, in predicting if a word belongs to a word class, is either true or false, and refers to the surrounding context: if = true, previous word = the otherwise The parameters are estimated by a procedure called Generalized Iterative Scaling (GIS) (Darroch and Ratcliff, 1972).","label":0},{"source_text":"Here we present two algorithms.","target_text":"For t = 1, T and for j = 1, 2: where 4 = exp(-jg'(xj,i)). practice, this greedy approach almost always results in an overall decrease in the value of Zco.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"7.96 5.55 1 l...................................................................................................................................................................................................J..","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This latter evaluation compares the performance of the system with that of several human judges since, as we shall show, even people do not agree on a single correct way to segment a text.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"4 53.7 43.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"A token that is allCaps will also be initCaps.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The complexity of the algorithm is O(E3 J2 2J), where E is the size of the target language vocabulary.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"As long as the main evaluation metric is dependency accuracy per word, with state-of-the-art accuracy mostly below 90%, the penalty for not handling non-projective constructions is almost negligible.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Finally, we model the probability of a new transliterated name as the product of PTN and PTN(hanzi;) for each hanzi; in the putative name.13 The foreign name model is implemented as an WFST, which is then summed with the WFST implementing the dictionary, morpho 13 The current model is too simplistic in several respects.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The PROBING model can perform optimistic searches by jumping to any n-gram without needing state and without any additional memory.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Recently, lattices have been used successfully in the parsing of Hebrew (Tsarfaty, 2006; Cohen and Smith, 2007), a Semitic language with similar properties to Arabic.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The observed performance gains, coupled with the simplicity of model implementation, makes it a compelling alternative to existing more complex counterparts.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This fact annoyed especially his dog...).","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"While there might be some controversy about the exact definition of such a tagset, these 12 categories cover the most frequent part-of-speech and exist in one form or another in all of the languages that we studied.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The computing time, the number of search errors, and the multi-reference WER (mWER) are shown as a function of t0.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"For a given partial hypothesis (C; j), the order in which the cities in C have been visited can be ignored (except j), only the score for the best path reaching j has to be stored.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In previous work (Bean and Riloff, 1999), we developed an unsupervised learning algorithm that automatically recognizes definite NPs that are existential without syntactic modification because their meaning is universally understood.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"(3)In sentence (1), McCann can be a person or an orga nization.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(Hearst 92) describes a method for extracting hyponyms from a corpus (pairs of words in &quot;isa&quot; relations).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Tables 4 and 5 also show that putting all of the contextual role KSs in play at the same time produces the greatest performance gain.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The inventions disclosed herein are the subject of a patent application owned by the University of Utah and licensed on an exclusive basis to Attensity Corporation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Denote by g3(x) = Et crithl(x) , j E {1,2} the unthresholded strong-hypothesis (i.e., f3 (x) = sign(gi (x))).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"However, this optimistic search would not visit the entries necessary to store backoff information in the outgoing state.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Moreover, if we want to test on a huge test corpus, indexing the whole corpus might prove computationally expensive.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"For each feature type f and tag t, a multinomial \u00cf\u0088tf is drawn from a symmetric Dirichlet distribution with concentration parameter \u00ce\u00b2.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Memory usage is the same as with binary search and lower than with set.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"For unaligned words, we set the tag to the most frequent tag in the corresponding treebank.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"These models generally outperform our memory consumption but are much slower, even when cached.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The corpus has been annotated with six different types of information, which are characterized in the following subsections.","label":1},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The ATB is disadvantaged by having fewer trees with longer average 5 LDC A-E catalog numbers: LDC2008E61 (ATBp1v4), LDC2008E62 (ATBp2v3), and LDC2008E22 (ATBp3v3.1).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":" We evaluate the system's performance by comparing its segmentation 'Tudgments\" with the judgments of a pool of human segmenters, and the system is shown to perform quite well.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"irL as the product of the probability estimate for i\u00c2\u00a5JJ1l., and the probability estimate just derived for unseen plurals in ir,: p(i\u00c2\u00a51J1l.ir,) p(i\u00c2\u00a51J1l.)p(unseen(f,)).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, there is a strong relationship between ni1s and the number of hanzi in the class.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The verbal reading arises when the maSdar has an NP argument which, in vocalized text, is marked in the accusative case.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"A sentence was withheld from section 22 because its extreme length was troublesome for a couple of the parsers.'","label":0},{"source_text":"Here we present two algorithms.","target_text":"Assume that the two classifiers are &quot;rote learners&quot;: that is, 1.1 and 12 are defined through look-up tables that list a label for each member of X1 or X2.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Dagan and Itai (Dagan and Itai, 1990) experimented with co-occurrence statistics that are similar to our lexical caseframe expectations.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"an event.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Supervised learning approaches have advanced the state-of-the-art on a variety of tasks in natural language processing, resulting in highly accurate systems.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Many researchers have developed coreference resolvers, so we will only discuss the methods that are most closely related to BABAR.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We map the ATB morphological analyses to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags for all experiments.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"This scenario is applicable to a large set of languages and has been considered by a number of authors in the past (Alshawi et al., 2000; Xi and Hwa, 2005; Ganchev et al., 2009).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Let us notate the set of previously unseen, or novel, members of a category X as unseen(X); thus, novel members of the set of words derived in f, menO will be de\u00c2\u00ad noted unseen(f,).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, if FCC and Federal Communications Commission are both found in a document, then Federal has A begin set to 1, Communications has A continue set to 1, Commission has A end set to 1, and FCC has A unique set to 1.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"In all of our experiments, the binary file (whether mapped or, in the case of most other packages, interpreted) is loaded into the disk cache in advance so that lazy mapping will never fault to disk.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"(f1; ;mg n fl1; l2; l3g ;m) German to English the monotonicity constraint is violated mainly with respect to the German verbgroup.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The content does not necessarily reflect the views of the U.S. Government, and no official endorsement should be inferred.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We will describe the evaluation of such clusters in the next subsection.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This is a straightforward technique that is arguably better suited to the adaptation task than the standard method of treating representative IN sentences as queries, then pooling the match results.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The Leaf Ancestor metric measures the cost of transforming guess trees to the reference (Sampson and Babarczy, 2003).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Top 5 Bot to m 5 Go ld NN P NN JJ CD NN S RB S PD T # \u00e2\u0080\u009d , 1T W CD W RB NN S VB N NN PR P$ W DT : MD . +P RI OR CD JJ NN S WP $ NN RR B- , $ \u00e2\u0080\u009d . +F EA TS JJ NN S CD NN P UH , PR P$ # . \u00e2\u0080\u009c Table 5: Type-level English POS Tag Ranking: We list the top 5 and bottom 5 POS tags in the lexicon and the predictions of our models under the best hyperparameter setting.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For example, a good classifier would identify Mrs. Frank as a person, Steptoe & Johnson as a company, and Honduras as a location.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We are interested in combining the substructures of the input parses to produce a better parse.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Thus, if one wants to segment words-for any purpose-from Chinese sentences, one faces a more difficult task than one does in English since one cannot use spacing as a guide.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We can only compare with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Some of the operations will be constant functions, corresponding to elementary structures, and will be written as f () = zi), where each z, is a constant, the string of terminal symbols al an,,,.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This class-based model gives reasonable results: for six radical classes, Table 1 gives the estimated cost for an unseen hanzi in the class occurring as the second hanzi in a double GIVEN name.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Several systems propose statistical methods for handling unknown words (Chang et al. 1992; Lin, Chiang, and Su 1993; Peng and Chang 1993).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"BABAR uses information extraction patterns to identify contextual roles and creates four contextual role knowledge sources using unsupervised learning.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"This group of features attempts to capture such information.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"These clusters are computed using an SVD variant without relying on transitional structure.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In the present study, we limit ourselves to an algorithmic approach, using a deterministic breadthfirst search.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Step 4.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We confirm the finding by Callison-Burch et al. (2006) that the rule-based system of Systran is not adequately appreciated by BLEU.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Can we do . QmS: Yes, wonderful.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Using the virtual distribution Di (i) and pseudo-labels&quot;y.,\u201e values for Wo, W\u00b1 and W_ can be calculated for each possible weak hypothesis (i.e., for each feature x E Xi); the weak hypothesis with minimal value for Wo + 2\/WW _ can be chosen as before; and the weight for this weak hypothesis at = ln ww+411:) can be calculated.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"They are also labelled for their topicality (yes \/ no), and this annotation is accompanied by a confidence value assigned by the annotator (since it is a more subjective matter).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We formulate the update as follows: where \u2200ui \u2208 Vf \\ Vfl, \u03b3i(y) and \u03bai are defined as: We ran this procedure for 10 iterations.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"This approach is similar to BABAR in that they both acquire knowledge from earlier resolutions.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"It was also proposed to allow annotators to skip sentences that they are unable to judge.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"NN \u00ef\u00bf\u00bd .e NP NNP NP DTNNP NN \u00ef\u00bf\u00bd .e NP NP NNP NP Table 5: Evaluation of 100 randomly sampled variation nuclei types.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The first modification \u2014 cautiousness \u2014 is a relatively minor change.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"This limits the number of NE category pairs to 2,000 and the number of NE pair instances to 0.63 million.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We have provided methods for handling certain classes of unknown words, and models for other classes could be provided, as we have noted.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"This suggests that different types of anaphora may warrant different treatment: definite NP resolution may depend more on lexical semantics, while pronoun resolution may depend more on contextual semantics.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The feature-based model replaces the emission distribution with a log-linear model, such that: on the word identity x, features checking whether x contains digits or hyphens, whether the first letter of x is upper case, and suffix features up to length 3.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Parent Head Modif er Dir # gold F1 Label # gold F1 NP NP TAG R 946 0.54 ADJP 1216 59.45 S S S R 708 0.57 SBAR 2918 69.81 NP NP ADJ P R 803 0.64 FRAG 254 72.87 NP NP N P R 2907 0.66 VP 5507 78.83 NP NP SBA R R 1035 0.67 S 6579 78.91 NP NP P P R 2713 0.67 PP 7516 80.93 VP TAG P P R 3230 0.80 NP 34025 84.95 NP NP TAG L 805 0.85 ADVP 1093 90.64 VP TAG SBA R R 772 0.86 WHN P 787 96.00 S VP N P L 961 0.87 (a) Major phrasal categories (b) Major POS categories (c) Ten lowest scoring (Collins, 2003)-style dependencies occurring more than 700 times Table 8: Per category performance of the Berkeley parser on sentence lengths \u00e2\u0089\u00a4 70 (dev set, gold segmentation).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We implement two data structures: PROBING, designed for speed, and TRIE, optimized for memory.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, as we have noted, nothing inherent in the approach precludes incorporating higher-order constraints, provided they can be effectively modeled within a finite-state framework.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The 8 similarity-to-IN features are based on word frequencies and scores from various models trained on the IN corpus: To avoid numerical problems, each feature was normalized by subtracting its mean and dividing by its standard deviation.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"5 67.3 55.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Then we ran binary search to determine the least amount of memory with which it would run.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We have not to date explored these various options.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"At first glance, the problem seems quite complex: a large number of rules is needed to cover the domain, suggesting that a large number of labeled examples is required to train an accurate classifier.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"\u00e2\u0080\u0094ested in correlations between prosody and dis course structure.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The belief value that would have been assigned to the intersection of these sets is .60*.70=.42, but this belief has nowhere to go because the null set is not permissible in the model.7 So this probability mass (.42) has to be redistributed.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In this section for the purposes of showing that polynomial time recognition is possible, we make the additional restriction that the contribution of a derived structure to the input string can be specified by a bounded sequence of substrings of the input.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Then, moving from connective to connective, ConAno sometimes offers suggestions for its scope (using heuristics like \u00e2\u0080\u0098for sub- junctor, mark all words up to the next comma as the first segment\u00e2\u0080\u0099), which the annotator can accept with a mouseclick or overwrite, marking instead the correct scope with the mouse.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We suspect that the higher precision in the disasters domain may be due to its substantially larger training corpus.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Three state-of-the-art statistical parsers are combined to produce more accurate parses, as well as new bounds on achievable Treebank parsing accuracy.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"kann 7.nicht 8.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Dropping the conditioning on 0 for brevity, and letting \u00afc\u03bb(s, t) = c\u03bb(s, t) + yu(s|t), and \u00afc\u03bb(t) = 4Note that the probabilities in (7) need only be evaluated over the support of \u02dcp(s, t), which is quite small when this distribution is derived from a dev set.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Our work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"For simplicity, we assume that OUT is homogeneous.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Find keywords for each NE pair The keywords are found for each NE category pair.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Because we are interested in applying our techniques to languages for which no labeled resources are available, we paid particular attention to minimize the number of free parameters and used the same hyperparameters for all language pairs.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The parallel data came from the Europarl corpus (Koehn, 2005) and the ODS United Nations dataset (UN, 2006).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Input: Ja , wunderbar . Konnen wir machen . MonS: Yes, wonderful.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The anaphor and antecedent appear in boldface.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In contrast, even though context can be helpful for resolving definite NPs, context can be trumped by the semantics of the nouns themselves.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"He also argues that paired English complementizers may also require structural descriptions whose path sets have nested dependencies.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Text generation, or at least the two phases of text planning and sentence planning, is a process driven partly by well-motivated choices (e.g., use this lexeme X rather than that more colloquial near-synonym Y ) and partly by con tation like that of PCC can be exploited to look for correlations in particular between syntactic structure, choice of referring expressions, and sentence-internal information structure.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Covering the first uncovered position in the source sentence, we use the language model probability p(ej$; $).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Even without features, but still using the tag prior, our median result is 52.0%, still significantly outperforming Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"). context=x The context for the entity.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"(Kehler, 1997) also used a DempsterShafer model to merge evidence from different sources for template-level coreference.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We also add an annotation for one-level iDafa (oneLevelIdafa) constructs since they make up more than 75% of the iDafa NPs in the ATB (Gabbard and Kulick, 2008).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Then, moving from connective to connective, ConAno sometimes offers suggestions for its scope (using heuristics like \u00e2\u0080\u0098for sub- junctor, mark all words up to the next comma as the first segment\u00e2\u0080\u0099), which the annotator can accept with a mouseclick or overwrite, marking instead the correct scope with the mouse.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Most similar is scrolling queries, wherein left-to-right queries that add one word at a time are optimized.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"An examination of the subjects' bracketings confirmed that these instructions were satisfactory in yielding plausible word-sized units.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The commentaries in PCC are all of roughly the same length, ranging from 8 to 10 sentences.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Some approaches depend upon some form of con\u00c2\u00ad straint satisfaction based on syntactic or semantic features (e.g., Yeh and Lee [1991], which uses a unification-based approach).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Moses keeps language models and many other resources in static variables, so these are still resident in memory.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"5 68.1 34.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"During coreference resolution, the caseframe network provides evidence that an anaphor and prior noun phrase might be coreferent.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The tree t2 must be on one of the two branches.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Step 2.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The state in future has not enough work for its many teachers.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"2.4 Underspecified rhetorical structure.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"However, the overall percentage of non-projective arcs is less than 2% in PDT and less than 1% in DDT.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"This vector tx is constructed for every word in the foreign vocabulary and will be used to provide features for the unsupervised foreign language POS tagger.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"However, the accuracy is considerably higher than previously reported results for robust non-projective parsing of Czech, with a best performance of 73% UAS (Holan, 2004).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Annotators argued for the importance of having correct and even multiple references.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The Berkeley parser gives state-of-the-art performance for all metrics.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"A few annotators suggested to break up long sentences into clauses and evaluate these separately.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We excluded these from the evaluation as they can be easily identified with a list of days\/months.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The Bikel GoldPOS configuration only supplies the gold POS tags; it does not force the parser to use them.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"He also argues that paired English complementizers may also require structural descriptions whose path sets have nested dependencies.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If they are found in a list, then a feature for that list will be set to 1.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"ni2ya3 and @5:2 xilya3, respectively.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In many cases, inspection of either the spelling or context alone is sufficient to classify an example.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Since MUC6, BBN' s Hidden Markov Model (HMM) based IdentiFinder (Bikel et al., 1997) has achieved remarkably good performance.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"3.1 General Knowledge Sources.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Other good classes include JADE and GOLD; other bad classes are DEATH and RAT.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The translation scores for the hypotheses generated with different threshold values t0 are compared to the translation scores obtained with a conservatively large threshold t0 = 10:0 . For each test series, we count the number of sentences whose score is worse than the corresponding score of the test series with the conservatively large threshold t0 = 10:0, and this number is reported as the number of search errors.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Decreasing the threshold results in higher mWER due to additional search errors.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Furthermore, by inverting the transducer so that it maps from phonemic transcriptions to hanzi sequences, one can apply the segmenter to other problems, such as speech recognition (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"A final alternate approach would be to combine weighted joint frequencies rather than conditional estimates, ie: cI(s, t) + w,\\(s, t)co(, s, t), suitably normalized.5 Such an approach could be simulated by a MAP-style combination in which separate 0(t) values were maintained for each t. This would make the model more powerful, but at the cost of having to learn to downweight OUT separately for each t, which we suspect would require more training data for reliable performance.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Tsarfaty (2006) used a morphological analyzer (Segal, 2000), a PoS tagger (Bar-Haim et al., 2005), and a general purpose parser (Schmid, 2000) in an integrated framework in which morphological and syntactic components interact to share information, leading to improved performance on the joint task.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"To define a similarity function between the English and the foreign vertices, we rely on high-confidence word alignments.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In figure 4, reverse relations are indicated by `*\u00e2\u0080\u0099 next to the frequency.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"At first glance, the problem seems quite complex: a large number of rules is needed to cover the domain, suggesting that a large number of labeled examples is required to train an accurate classifier.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"For other languages, we use the CoNLL-X multilingual dependency parsing shared task corpora (Buchholz and Marsi, 2006) which include gold POS tags (used for evaluation).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The Danish Dependency Treebank (DDT) comprises about 100K words of text selected from the Danish PAROLE corpus, with annotation of primary and secondary dependencies (Kromann, 2003).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Tokens tagged as PUNC are not discarded unless they consist entirely of punctuation.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"One conclusion drawn from this annotation effort was that for humans and machines alike, 2 www.sfs.nphil.unituebingen.de\/Elwis\/stts\/ stts.html 3 www.coli.unisb.de\/sfb378\/negra-corpus\/annotate.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Our oracles took advantage of the labeled treebanks: While we tried to minimize the number of free parameters in our model, there are a few hyperparameters that need to be set.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"na me =>1 ha nzi fa mi ly 1 ha nzi gi ve n 4.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The label for a test example with features x is then defined as In this paper we define h(x, y) as the following function of counts seen in training data: Count(x,y) is the number of times feature x is seen with label y in training data, Count(x) = EyEy Count(x, y). a is a smoothing parameter, and k is the number of possible labels.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Let us consider an example of ambiguity caused by devocalization.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Less frequently studied is the interplay among language, annotation choices, and parsing model design (Levy and Manning, 2003; Ku\u00c2\u00a8 bler, 2005).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We suggest that in unlexicalized PCFGs the syntactic context may be explicitly modeled in the derivation probabilities.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Also there are cases where one of the two NEs belong to a phrase outside of the relation.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Figure 4 shows a constituent headed by a process nominal with an embedded adjective phrase.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"These universal POS categories not only facilitate the transfer of POS information from one language to another, but also relieve us from using controversial evaluation metrics,2 by establishing a direct correspondence between the induced hidden states in the foreign language and the observed English labels.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The first method uses a similar algorithm to that of (Yarowsky 95), with modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"We group the features used into feature groups.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This has the effect of randomly permuting vocabulary identifiers, meeting the requirements of interpolation search when vocabulary identifiers are used as keys.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"49 99.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"1 53.8 47.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This extends previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and using a simpler training procedure.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"5.2 Setup.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"This number must be less than or equal to n \u00f4\u0080\u0080\u0080 1.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We experimented with increasingly rich grammars read off of the treebank.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We would like to thank Ryan McDonald for numerous discussions on this topic.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"This procedure is repeated for T rounds while alternating between the two classifiers.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"These 140 NE categories are designed by extending MUC\u00e2\u0080\u0099s 7 NE categories with finer sub-categories (such as Company, Institute, and Political Party for Organization; and Country, Province, and City for Location) and adding some new types of NE categories (Position Title, Product, Event, and Natural Object).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"5 70.1 58.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In this case nonalpha is the string formed by removing all upper\/lower case letters from the spelling (e.g., for Thomas E. Petry nonalpha= .","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The knowledge base then can be tested for its relation-inference capabilities on the basis of full-blown co-reference information.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"11 taTweel (-) is an elongation character used in Arabic script to justify text.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The secondary classifier in (Borthwick, 1999) uses information not just from the current article, but also from the whole test corpus, with an additional feature that indicates if the information comes from the same document or from another document.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Translation errors are reported in terms of multireference word error rate (mWER) and subjective sentence error rate (SSER).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In sequential tagging models such as (Adler and Elhadad, 2006; Bar-Haim et al., 2007; Smith et al., 2005) weights are assigned according to a language model The input for the joint task is a sequence W = w1, ... , wn of space-delimited tokens.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The corresponding token words w are drawn conditioned on t and \u00ce\u00b8.2 Our full generative model is given by: K P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) = n (P (\u00cf\u0086t|\u00ce\u00b1)P (\u00ce\u00b8t|T , \u00ce\u00b1)) t=1 The transition distribution \u00cf\u0086t for each tag t is drawn according to DIRICHLET(\u00ce\u00b1, K ), where \u00ce\u00b1 is the shared transition and emission distribution hyperparameter.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For a word whose initCaps might be due to its position rather than its meaning (in headlines, first word of a sentence, etc), the case information of other occurrences might be more accurate than its own.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The details are given in (Tillmann, 2000).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"3.1 Maximum Entropy.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Step 1.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"This approach leads to a search procedure with complexity O(E3 J4).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"5.1 Parsing Models.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Among the IS-units, the referring expressions are marked as such and will in the second phase receive a label for cognitive status (active, accessible- text, accessible-situation, inferrable, inactive).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Before presenting our results, we describe the datasets that we used, as well as two baselines.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"To bridge this gap, we consider a practically motivated scenario, in which we want to leverage existing resources from a resource-rich language (like English) when building tools for resource-poor foreign languages.1 We assume that absolutely no labeled training data is available for the foreign language of interest, but that we have access to parallel data with a resource-rich language.","label":0},{"source_text":"There is no global pruning.","target_text":"mein 5.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"orthographic words are thus only a starting point for further analysis and can only be regarded as a useful hint at the desired division of the sentence into words.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The state function is integrated into the query process so that, in lieu of the query p(wnjwn\u22121 1 ), the application issues query p(wnjs(wn\u22121 1 )) which also returns s(wn1 ).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"92 76.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The form mnh itself can be read as at least three different verbs (\u201ccounted\u201d, \u201cappointed\u201d, \u201cwas appointed\u201d), a noun (\u201ca portion\u201d), and a possessed noun (\u201cher kind\u201d).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"While processing the source sentence monotonically, the initial state I is entered whenever there are no uncovered positions to the left of the rightmost covered position.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"However there is no global pruning.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"yu2 'fish.'","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Note that in our formalism a weakhypothesis can abstain.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"In general, the lemma of the previous section does not ensure that all the productions in the combined parse are found in the grammars of the member parsers.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The zone to which a token belongs is used as a feature.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Once HMM parameters (\u00ce\u00b8, \u00cf\u0086) are drawn, a token-level tag and word sequence, (t, w), is generated in the standard HMM fashion: a tag sequence t is generated from \u00cf\u0086.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Furthermore, we know one of the original parses will be the hypothesized parse, so the direct method of determining which one is best is to compute the probability of each of the candidate parses using the probabilistic model we developed in Section 2.1.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The set is then compared with the set generated from the Penn Treebank parse to determine the precision and recall.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We realize the importance of paraphrase; however, the major obstacle is the construction of paraphrase knowledge.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Restrictions: Quasi-monotone Search The above search space is still too large to allow the translation of a medium length input sentence.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The only way to handle such phenomena within the framework described here is simply to expand out the reduplicated forms beforehand, and incorporate the expanded forms into the lexical transducer.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The baseline system in Table 3 refers to the maximum entropy system that uses only local features.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"All the sentences have been analyzed by our chunker and NE tag- ger.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Bikel 75 training trees 5000 10000 15000 Figure 3: Dev set learning curves for sentence lengths \u00e2\u0089\u00a4 70.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Each trie node contains a sorted array of entries and they use binary search.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In our experiment, we set the threshold of the TF\/ITF score empirically using a small development corpus; a finer adjustment of the threshold could reduce the number of such keywords.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"com \u00c2\u00a7Cambridge, UK Email: nc201@eng.cam.ac.uk \u00c2\u00a9 1996 Association for Computational Linguistics (a) B ) ( , : & ; ? ' H o w d o y o u s a y o c t o p u s i n J a p a n e s e ? ' (b) P l a u s i b l e S e g m e n t a t i o n I B X I I 1 : & I 0 0 r i 4 w e n 2 z h a n g l y u 2 z e n 3 m e 0 s h u o l ' J a p a n e s e ' ' o c t o p u s ' ' h o w ' ' s a y ' (c) Figure 1 I m p l a u s i b l e S e g m e n t a t i o n [\u00c2\u00a7] lxI 1:&I ri4 wen2 zhangl yu2zen3 me0 shuol 'Japan' 'essay' 'fish' 'how' 'say' A Chinese sentence in (a) illustrating the lack of word boundaries.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The translation search is carried out with the category markers and the city names are resubstituted into the target sentence as a postprocessing step.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Finally, the statistical method fails to correctly group hanzi in cases where the individual hanzi comprising the name are listed in the dictionary as being relatively high-frequency single-hanzi words.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"However, reads in the TRIE data structure are more expensive due to bit-level packing, so we found that it is faster to use interpolation search the entire time.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(1999) did not report using any dictionaries, but mentioned in a footnote that they have added list membership features, which have helped marginally in certain domains.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Both of the switching techniques, as well as the parametric hybridization technique were also shown to be robust when a poor parser was introduced into the experiments.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The hash variant is a reverse trie with hash tables, a more memory-efficient version of SRILM\u2019s default.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The resulting structural differences between tree- banks can account for relative differences in parsing performance.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Since trees in a tree set are adjoined together, the addressing scheme uses a sequence of pairings of the address and name of the elementary tree adjoined at that address.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Other packages walk their respective data structures once to find wnf and again to find {b(wn\u22121 i )}f\u22121 i=1if necessary.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Consider the case where IX].","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Despite these limitations, a purely finite-state approach to Chinese word segmentation enjoys a number of strong advantages.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Human evaluation is one way to distinguish between the two cases.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Our system fails in (a) because of$ shenl, a rare family name; the system identifies it as a family name, whereas it should be analyzed as part of the given name.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The problem is a binary classification problem.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Evaluation results for links","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"\u00c2\u00a3 : _ADV: 5.88 If:!","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Reading the following record\u2019s offset indicates where the block ends.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The rationale for treating these semantic labels differently is that they are specific and reliable (as opposed to the WordNet classes, which are more coarse and more noisy due to polysemy).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We first note that the accuracy results of our system are overall higher on their setup, on all measures, indicating that theirs may be an easier dataset.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"2 We now introduce a new algorithm for learning from unlabeled examples, which we will call DLCoTrain (DL stands for decision list, the term Cotrain is taken from (Blum and Mitchell 98)).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"As a result, Arabic sentences are usually long relative to English, especially after Length English (WSJ) Arabic (ATB) \u00e2\u0089\u00a4 20 41.9% 33.7% \u00e2\u0089\u00a4 40 92.4% 73.2% \u00e2\u0089\u00a4 63 99.7% 92.6% \u00e2\u0089\u00a4 70 99.9% 94.9% Table 2: Frequency distribution for sentence lengths in the WSJ (sections 2\u00e2\u0080\u009323) and the ATB (p1\u00e2\u0080\u00933).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For derived words that occur in our corpus we can estimate these costs as we would the costs for an underived dictionary entry.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"A k-tape ATM, M, has a read-only input tape and k read-write work tapes.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In the cotraining case, (Blum and Mitchell 98) argue that the task should be to induce functions Ii and f2 such that So Ii and 12 must (1) correctly classify the labeled examples, and (2) must agree with each other on the unlabeled examples.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In (Bikel et al., 1997) and (Bikel et al., 1999), performance was plotted against training data size to show how performance improves with training data size.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"5 \u00e2\u0080\u0098Underspecified Rhetorical Markup Language\u00e2\u0080\u0099 6 This confirms the figure given by (Schauer, Hahn.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"2.2.4 Semantic Caseframe Expectations The third type of contextual role knowledge learned by BABAR is Semantic Caseframe Expectations.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Several systems propose statistical methods for handling unknown words (Chang et al. 1992; Lin, Chiang, and Su 1993; Peng and Chang 1993).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If is seen infrequently during training (less than a small count), then will not be selected as a feature and all features in this group are set to 0.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The same argument holds for resolving PP attachment of a prefixed preposition or marking conjunction of elements of any kind.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Following the convention presented in earlier sections, we assume that each example is an instance pair of the from (xi ,i, x2,) where xj,, E 2x3 , j E 2}.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Hence, their relationship to formalisms such as HG's and TAG's is of interest.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"An input ABCD can be represented as an FSA as shown in Figure 2(b).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The motivation for our more informal approach was the intuition that there are so many open problems in rhetorical analysis (and more so for German than for English; see below) that the main task is qualitative investigation, whereas rigorous quantitative analyses should be performed at a later stage.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"However, we do not explore this possibility in the current work.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"so that 'door' would be and in this case the hanzi 7C, does not represent a syllable.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The segmentation chosen is the best path through the WFST, shown in (d).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Step 4.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"A final alternate approach would be to combine weighted joint frequencies rather than conditional estimates, ie: cI(s, t) + w,\\(s, t)co(, s, t), suitably normalized.5 Such an approach could be simulated by a MAP-style combination in which separate 0(t) values were maintained for each t. This would make the model more powerful, but at the cost of having to learn to downweight OUT separately for each t, which we suspect would require more training data for reliable performance.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"It was developed in response to the non-terminal\/terminal bias of Evalb, but Clegg and Shepherd (2005) showed that it is also a valuable diagnostic tool for trees with complex deep structures such as those found in the ATB.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Different annotations of the same text are mapped into the same data structure, so that search queries can be formulated across annotation levels.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Turning now to (1), we have the similar problem that splitting.into.ma3 'horse' andlu4 'way' is more costly than retaining this as one word .ma3lu4 'road.'","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"As indicated by bolding, for seven out of eight languages the improvements of the \u201cWith LP\u201d setting are statistically significant with respect to the other models, including the \u201cNo LP\u201d setting.11 Overall, it performs 10.4% better than the hitherto state-of-the-art feature-HMM baseline, and 4.6% better than direct projection, when we macro-average the accuracy over all languages.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For example, as Gan (1994) has noted, one can construct examples where the segmen\u00c2\u00ad tation is locally ambiguous but can be determined on the basis of sentential or even discourse context.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The key to the methods we describe is redundancy in the unlabeled data.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The following two error criteria are used in our experiments: mWER: multi-reference WER: We use the Levenshtein distance between the automatic translation and several reference translations as a measure of the translation errors.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Fig.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(b) F.i'JJI!","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"VBD she added VP PUNC \u00e2\u0080\u009c SBAR IN NP 0 NN.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Each vertex within a connected component must have the same label \u2014 in the binary classification case, we need a single labeled example to identify which component should get which label.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For humans, this characteristic can impede the acquisition of literacy.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The orthographic normalization strategy we use is simple.10 In addition to removing all diacritics, we strip instances of taTweel J=J4.i, collapse variants of alif to bare alif,11 and map Ara bic punctuation characters to their Latin equivalents.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The standard measures for evaluating Penn Treebank parsing performance are precision and recall of the predicted constituents.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"64 76.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Two general approaches are presented and two combination techniques are described for each approach.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"However, their inverted variant implements a reverse trie using less CPU and the same amount of memory7.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In our coreference resolver, we define \u00ce\u00b8 to be the set of all candidate antecedents for an anaphor.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"(2009).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The effect of a second reference resolution classifier is not entirely the same as that of global features.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"If they knew that the first four words in a hypergraph node would never extend to the left and form a 5-gram, then three or even fewer words could be kept in the backward state.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The theory has also been validated empirically.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We used the MUC4 terrorism corpus (MUC4 Proceedings, 1992) and news articles from the Reuter\u00e2\u0080\u0099s text collection8 that had a subject code corresponding to natural disasters.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"In general, the neighborhoods can be more diverse and we allow a soft label distribution over the vertices.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(2) was extended to have an additional, innermost loop over the (3) possible labels.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"To quantize, we use the binning method (Federico and Bertoldi, 2006) that sorts values, divides into equally sized bins, and averages within each bin.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"These models generally outperform our memory consumption but are much slower, even when cached.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Jud ges A G G R ST M 1 M 2 M 3 T1 T2 T3 AG 0.7 0 0.7 0 0 . 4 3 0.4 2 0.6 0 0.6 0 0.6 2 0.5 9 GR 0.9 9 0 . 6 2 0.6 4 0.7 9 0.8 2 0.8 1 0.7 2 ST 0 . 6 4 0.6 7 0.8 0 0.8 4 0.8 2 0.7 4 M1 0.7 7 0.6 9 0.7 1 0.6 9 0.7 0 M2 0.7 2 0.7 3 0.7 1 0.7 0 M3 0.8 9 0.8 7 0.8 0 T1 0.8 8 0.8 2 T2 0.7 8 respectively, the recall and precision.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The learned patterns are then normalized and applied to the corpus.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In this section we present a partial evaluation of the current system, in three parts.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(1992).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"What both of these approaches presume is that there is a sin\u00c2\u00ad gle correct segmentation for a sentence, against which an automatic algorithm can be compared.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"4.1 Corpora.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Semantic expectations are analogous to lexical expectations except that they represent semantic classes rather than nouns.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Given names are most commonly two hanzi long, occasionally one hanzi long: there are thus four possible name types, which can be described by a simple set of context-free rewrite rules such as the following: 1.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Although we have not done any experiments on other languages, this way of using global features from a whole document should be applicable to other languages.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Timing is based on plentiful memory.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Each set is assigned two values: belief and plausibility.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"While RST (Mann, Thompson 1988) proposed that a single relation hold between adjacent text segments, SDRT (Asher, Lascarides 2003) maintains that multiple relations may hold simultaneously.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"then define the best segmentation to be the cheapest or best path in Id(I) o D* (i.e., Id(I) composed with the transitive closure of 0).6 Consider the abstract example illustrated in Figure 2.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Note that it is in precision that our over\u00c2\u00ad all performance would appear to be poorer than the reported performance of Chang et al., yet based on their published examples, our system appears to be doing better precisionwise.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Applying the function PROJECTIVIZE to the graph in Figure 1 yields the graph in Figure 2, where the problematic arc pointing to Z has been lifted from the original head jedna to the ancestor je.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The similar explanation applies to the link to the \u00e2\u0080\u009cstake\u00e2\u0080\u009d set.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This leads to a linear combination of domain-specific probabilities, with weights in [0, 1], normalized to sum to 1.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The idea is to have a pipeline of shallow-analysis modules (tagging, chunk- ing, discourse parsing based on connectives) and map the resulting underspecified rhetorical tree (see Section 2.4) into a knowledge base that may contain domain and world knowledge for enriching the representation, e.g., to resolve references that cannot be handled by shallow methods, or to hypothesize coherence relations.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":", Sun day, then the feature DayOfTheWeek is set to 1.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"KS Function Ge nde r filters candidate if gender doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We further thank Khalil Simaan (ILLCUvA) for his careful advise concerning the formal details of the proposal.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"to explore how well we can induce POS tags using only the one-tag-per-word constraint.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"0 55.3 34.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In total, across all domains, we kept 13,976 phrases with keywords.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Each parse is converted into a set of constituents represented as a tuples: (label, start, end).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"genitiveMark indicates recursive NPs with a indefinite nominal left daughter and an NP right daughter.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"4.5 Transliterations of Foreign Words.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Feature weights were set using Och\u2019s MERT algorithm (Och, 2003).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"F-measure is the harmonic mean of precision and recall, 2PR\/(P + R).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Many packages perform language model queries.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Similarly, hanzi sharing the GHOST radical _m tend to denote spirits and demons, such as _m gui3 'ghost' itself, II: mo2 'demon,' and yan3 'nightmare.'","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We plan to explore more powerful techniques for exploiting the diversity of parsing methods.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Covering the first uncovered position in the source sentence, we use the language model probability p(ej$; $).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Obviously \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is part of the following clause rather than being the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d and the extracted instance makes no sense.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"So, it is too costly to make IE technology \u00e2\u0080\u009copen- domain\u00e2\u0080\u009d or \u00e2\u0080\u009con-demand\u00e2\u0080\u009d like IR or QA.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"However, lazy mapping is generally slow because queries against uncached pages must wait for the disk.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In the appositive case, the contextual predictor was the head of the modifying appositive (president in the Maury Cooper example); in the second case, the contextual predictor was the preposition together with the noun it modifies (plant_in in the Georgia example).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"By making use of global context, it has achieved excellent results on both MUC6 and MUC7 official test data.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The segmenter will give both analyses :1 cai2 neng2 'just be able,' and ?]cai2neng2 'talent,' but the latter analysis is preferred since splitting these two morphemes is generally more costly than grouping them.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Before presenting our results, we describe the datasets that we used, as well as two baselines.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This limits the number of NE category pairs to 2,000 and the number of NE pair instances to 0.63 million.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Zone: MUC data contains SGML tags, and a document is divided into zones (e.g., headlines and text zones).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"na me =>2 ha nzi fa mi ly 2 ha nzi gi ve n 5.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"IRSTLM 5.60.02 (Federico et al., 2008) is a sorted trie implementation designed for lower memory consumption.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"3.1 Lexicon Component.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Since every CFL is known to be semilinear (Parikh, 1966), in order to show semilinearity of some language, we need only show the existence of a letter equivalent CFL Our definition of LCFRS's insists that the composition operations are linear and nonerasing.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"level.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Unlike Germann et al. (2009), we chose a model size so that all benchmarks fit comfortably in main memory.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The cost estimate, cost(i\u00c2\u00a5JJ1l.fn is computed in the obvious way by summing the negative log probabilities of i\u00c2\u00a5JJ1l.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"However, TRIE partitions storage by n-gram length, so walking the trie reads N disjoint pages.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"There are two possible reasons: (1) the knowledge sources are resolving different cases of anaphora, and (2) the knowledge sources provide multiple pieces of evidence in support of (or against) a candidate, thereby acting synergistically to push the DempsterShafer model over the belief threshold in favor of a single candidate.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The \u201cNo LP\u201d model does not outperform direct projection for German and Greek, but performs better for six out of eight languages.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Finally, we thank Kuzman Ganchev and the three anonymous reviewers for helpful suggestions and comments on earlier drafts of this paper.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"On the other hand, only very restricted reorderings are necessary, e.g. for the translation direction from Table 2: Coverage set hypothesis extensions for the IBM reordering.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The compact variant uses sorted arrays instead of hash tables within each node, saving some memory, but still stores full 64-bit pointers.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"It is sometimes claimed that one of the advantages of dependency grammar over approaches based on constituency is that it allows a more adequate treatment of languages with variable word order, where discontinuous syntactic constructions are more common than in languages like English (Mel\u2019\u02c7cuk, 1988; Covington, 1990).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Sheffield Guthrie and Hepple (2010) explore several randomized compression techniques, but did not release code.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"52 77.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"For all languages we do not make use of a tagging dictionary.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"C is the union of the sets of constituents suggested by the parsers. r(c) is a binary function returning t (for true) precisely when the constituent c E C should be included in the hypothesis.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"For illustration, an English translation of one of the commentaries is given in Figure 1.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For a sentence x and a state sequence z, a first order Markov model defines a distribution: (9) where Val(X) corresponds to the entire vocabulary.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"When the signal is a coordinating conjunction, the second span is usually the clause following the conjunction; the first span is often the clause preceding it, but sometimes stretches further back.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We used this data to build an unpruned ARPA file with IRSTLM\u2019s improved-kneser-ney option and the default three pieces.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"29 \u00e2\u0080\u0094 95.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"They are set to fixed constants.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For all variants, we found that BerkeleyLM always rounds the floating-point mantissa to 12 bits then stores indices to unique rounded floats.","label":0},{"source_text":"There is no global pruning.","target_text":"In Section 4, we present the performance measures used and give translation results on the Verbmobil task.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Named Entity Recognition: A Maximum Entropy Approach Using Global Information","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For speed, we plan to implement the direct-mapped cache from BerkeleyLM.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"M spawns as many processes as there are ways of breaking up ri , .. \u2022 , zt, and rules with A on their left-hand-side.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Even if an example like this is not labeled, it can be interpreted as a &quot;hint&quot; that Mr and president imply the same category.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Features and context were initially introduced into the models, but they refused to offer any gains in performance.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The tree t2 must be on one of the two branches.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If all three sentences are in the same document, then even a human will find it difficult to classify McCann in (1) into either person or organization, unless there is some other information provided.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thus, any language that is letter equivalent to a semilinear language is also semilinear.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"In all of our experiments, the binary file (whether mapped or, in the case of most other packages, interpreted) is loaded into the disk cache in advance so that lazy mapping will never fault to disk.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(1) CEO of McCann . . .","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"and f,.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The computing time is low, since no reordering is carried out.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Both (Tsarfaty, 2006; Cohen and Smith, 2007) have shown that a single integrated framework outperforms a completely streamlined implementation, yet neither has shown a single generative model which handles both tasks.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"5 \u00e2\u0080\u0098Underspecified Rhetorical Markup Language\u00e2\u0080\u0099 6 This confirms the figure given by (Schauer, Hahn.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Statistics for all data sets are shown in Table 2.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Fourth, we show how to build better models for three different parsers.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Also, expanding on the techniques for the automatic generation of extraction patterns (Riloff 96; Sudo 03) using our method, the extraction patterns which have the same meaning can be automatically linked, enabling us to produce the final table fully automatically.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Terrorism systems must distinguish between people who perpetrate a crime and people who are victims of a crime.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The content does not necessarily reflect the views of the U.S. Government, and no official endorsement should be inferred.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Overall, language modeling significantly impacts decoder performance.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Also, the method of using keywords rules out phrases which don\u00e2\u0080\u0099t contain popular words in the domain.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"A large number of rules is needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classi- However, we show that the use of data can reduce the requirements for supervision to just 7 simple &quot;seed&quot; rules.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Because there might be some controversy about the exact definitions of such universals, this set of coarse-grained POS categories is defined operationally, by collapsing language (or treebank) specific distinctions to a set of categories that exists across all languages.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"60 ( P e tr o v, 2 0 0 9 ) all B e r k e l e y ( S e p . 0 9 ) B a s e l i n e 7 0 a l l G o l d P O S 70 \u00e2\u0080\u0094 \u00e2\u0080\u0094 \u00e2\u0080\u0094 0 . 8 0 9 0.839 335 0 . 7 9","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The POS distributions over the foreign trigram types are used as features to learn a better unsupervised POS tagger (\u00a75).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In this specific case, as these two titles could fill the same column of an IE table, we regarded them as paraphrases for the evaluation.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"As suggested at the end of Section 3, the restrictions that have been specified in the definition of LCFRS's suggest that they can be efficiently recognized.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"32 81.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We applied the AutoSlog system (Riloff, 1996) to our unannotated training texts to generate a set of extraction patterns for each domain.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"If we remove this sample from the evaluation, then the ATB type-level error rises to only 37.4% while the n-gram error rate increases to 6.24%.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The precision and recall of similarity switching and constituent voting are both significantly better than the best individual parser, and constituent voting is significantly better than parser switching in precision.4 Constituent voting gives the highest accuracy for parsing the Penn Treebank reported to date.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"One way to approach this discrepancy is to assume a preceding phase of morphological segmentation for extracting the different lexical items that exist at the token level (as is done, to the best of our knowledge, in all parsing related work on Arabic and its dialects (Chiang et al., 2006)).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Notice that even though IG's and LFG's involve CFG-like productions, they are (linguistically) fundamentally different from CFG's because the composition operations need not be linear.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"To provide a thorough analysis, we evaluated three baselines and two oracles in addition to two variants of our graph-based approach.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As the reviewer also points out, this is a problem that is shared by, e.g., probabilistic context-free parsers, which tend to pick trees with fewer nodes.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Past work however, has typically associ n = n P (Ti)P (Wi|Ti) = i=1 1 n K n ated these features with token occurrences, typically in an HMM.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"96 75.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For example, the Wang, Li, and Chang system fails on the sequence 1:f:p:]nian2 nei4 sa3 in (k) since 1F nian2 is a possible, but rare, family name, which also happens to be written the same as the very common word meaning 'year.'","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The kind of annotation work presented here would clearly benefit from the emergence of standard formats and tag sets, which could lead to sharable resources of larger size.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, it is possible to personify any noun, so in children's stories or fables, i\u00c2\u00a5JJ1l.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Other errors include NE tagging errors and errors due to a phrase which includes other NEs.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"This group of features attempts to capture such information.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The first stage identifies a keyword in each phrase and joins phrases with the same keyword into sets.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"This paper proposes a simple and effective tagging method that directly models tag sparsity and other distributional properties of valid POS tag assignments.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"To initialize the graph for label propagation we use a supervised English tagger to label the English side of the bitext.7 We then simply count the individual labels of the English tokens and normalize the counts to produce tag distributions over English word types.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thus, the derivation trees for TAG's have the same structure as local sets.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"This section describes an algorithm based on boosting algorithms, which were previously developed for supervised machine learning problems.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Despite their simplicity, uni- gram weights have been shown as an effective feature in segmentation models (Dyer, 2009).13 The joint parser\/segmenter is compared to a pipeline that uses MADA (v3.0), a state-of-the-art Arabic segmenter, configured to replicate ATB segmentation (Habash and Rambow, 2005).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"While the proportion of sentences containing non-projective dependencies is often 15\u201325%, the total proportion of non-projective arcs is normally only 1\u20132%.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"This process is repeated 5 times by rotating the data appropriately.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"For example, syntactic decoders (Koehn et al., 2007; Dyer et al., 2010; Li et al., 2009) perform dynamic programming parametrized by both backward- and forward-looking state.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"then define the best segmentation to be the cheapest or best path in Id(I) o D* (i.e., Id(I) composed with the transitive closure of 0).6 Consider the abstract example illustrated in Figure 2.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"besuchen 9.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The intuition here is that the role of a discourse marker can usually be de 9 Both the corpus split and pre-processing code are avail-.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"When SRILM estimates a model, it sometimes removes n-grams but not n + 1-grams that extend it to the left.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Consider the case where IX].","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Surprisingly, the non-parametric switching technique also exhibited robust behaviour in this situation.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Their theoretical finding is simply stated: classification error rate decreases toward the noise rate exponentially in the number of independent, accurate classifiers.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"A crucial difference is that the number of parameters is greatly reduced as is the number of variables that are sampled during each iteration.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Here, annotation proceeds in two phases: first, the domains and the units of IS are marked as such.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"4 Traditional Arabic linguistic theory treats both of these types as subcategories of noun \u00ef\u00bf\u00bd '.i . Figure 1: The Stanford parser (Klein and Manning, 2002) is unable to recover the verbal reading of the unvocalized surface form 0 an (Table 1).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In this domain the major scenarios involve the things they agreed on, rather than the mere fact that they agreed.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Our work is motivated by the observation that contextual roles can be critically important in determining the referent of a noun phrase.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Statistics for all data sets are shown in Table 2.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"As one can see, only the trigrams [suo incarceramento ,], [suo iter ,] and [suo carattere ,] are connected to English words.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"shortest match at each point.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As we have seen, the lexicon of basic words and stems is represented as a WFST; most arcs in this WFST represent mappings between hanzi and pronunciations, and are costless.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We checked whether the discovered links are listed in WordNet.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We can 5 Recall that precision is defined to be the number of correct hits divided by the total number of items.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Any NLP application that presumes as input unrestricted text requires an initial phase of text analysis; such applications involve problems as diverse as machine translation, information retrieval, and text-to-speech synthesis (TIS).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"9 www.ling.unipotsdam.de\/sfb\/ Figure 2: Screenshot of Annis Linguistic Database 3.3 Symbolic and knowledge-based.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The most frequent NE category pairs are \u00e2\u0080\u009cPerson - Person (209,236), followed by \u00e2\u0080\u009cCountry - Coun- try\u00e2\u0080\u009d (95,123) and \u00e2\u0080\u009cPerson - Country\u00e2\u0080\u009d (75,509).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We have shown that, at least given independent human judgments, this is not the case, and that therefore such simplistic measures should be mistrusted.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Finally, we note that Jiang\u2019s instance-weighting framework is broader than we have presented above, encompassing among other possibilities the use of unlabelled IN data, which is applicable to SMT settings where source-only IN corpora are available.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"2 62.2 39.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Restrictions: Quasi-monotone Search The above search space is still too large to allow the translation of a medium length input sentence.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Of course, since the number of attested (phonemic) Mandarin syllables (roughly 1400, including tonal distinctions) is far smaller than the number of morphemes, it follows that a given syllable could in principle be written with any of several different hanzi, depending upon which morpheme is intended: the syllable zhongl could be lfl 'middle,''clock,''end,' or ,'loyal.'","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"0750271 and by the DARPA GALE program.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The translation of one position in the source sentence may be postponed for up to L = 3 source positions, and the translation of up to two source positions may be anticipated for at most R = 10 source positions.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The range of the score is between 0 and 1 (higher is better).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Third, we develop a human interpretable grammar that is competitive with a latent variable PCFG.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"A procedural definition to restrict1In the approach described in (Berger et al., 1996), a mor phological analysis is carried out and word morphemes rather than full-form words are used during the search.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Nonetheless, the results of the comparison with human judges demonstrates that there is mileage being gained by incorporating models of these types of words.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We indicate whether a context with zero log backoff will extend using the sign bit: +0.0 for contexts that extend and \u22120.0 for contexts that do not extend.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The algorithm, called CoBoost, has the advantage of being more general than the decision-list learning alInput: (xi , yi), , (xim, ) ; x, E 2x, yi = +1 Initialize Di (i) = 1\/m.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"There is a fairly large body of work on SMT adaptation.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"To see this, note thai the first two terms in the above equation correspond to the function that AdaBoost attempts to minimize in the standard supervised setting (Equ.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"We would like to thank Eugene Charniak, Michael Collins, and Adwait Ratnaparkhi for enabling all of this research by providing us with their parsers and helpful comments.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The difference in precision between similarity and Bayes switching techniques is significant, but the difference in recall is not.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"2.6 Co-reference.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"10 Other orthographic normalization schemes have been suggested for Arabic (Habash and Sadat, 2006), but we observe negligible parsing performance differences between these and the simple scheme used in this evaluation.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For monolingual treebank data we relied on the CoNLL-X and CoNLL-2007 shared tasks on dependency parsing (Buchholz and Marsi, 2006; Nivre et al., 2007).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The Wang, Li, and Chang system fails on fragment (b) because their system lacks the word youlyoul 'soberly' and misinterpreted the thus isolated first youl as being the final hanzi of the preceding name; similarly our system failed in fragment (h) since it is missing the abbreviation i:lJI!","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For example, the Wang, Li, and Chang system fails on the sequence 1:f:p:]nian2 nei4 sa3 in (k) since 1F nian2 is a possible, but rare, family name, which also happens to be written the same as the very common word meaning 'year.'","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Our TRIE implements the popular reverse trie, in which the last word of an n-gram is looked up first, as do SRILM, IRSTLM\u2019s inverted variant, and BerkeleyLM except for the scrolling variant.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The high 1 tone of J1l would not normally neutralize in this fashion if it were functioning as a word on its own.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"2.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The probability distribution that satisfies the above property is the one with the highest entropy.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The percentage scores on the axis labels represent the amount of variation in the data explained by the dimension in question.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"There are two weaknesses in Chang et al.'s model, which we improve upon.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"To approximate these baselines, we implemented a very simple sentence selection algorithm in which parallel sentence pairs from OUT are ranked by the perplexity of their target half according to the IN language model.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"There has been additional recent work on inducing lexicons or other knowledge sources from large corpora.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In the CC-domain, there are 32 sets of phrases which contain more than 2 phrases.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In the CC-domain, there are 32 sets of phrases which contain more than 2 phrases.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The system described in this paper is similar to the MENE system of (Borthwick, 1999).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For all these annotation taks, G\u00c2\u00a8otze developed a series of questions (essentially a decision tree) designed to lead the annotator to the ap propriate judgement.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"tai2du2 'Taiwan Independence.'","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This section measures performance on shared tasks in order of increasing complexity: sparse lookups, evaluating perplexity of a large file, and translation with Moses.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Since our destructor is an efficient call to munmap, bypassing the destructor favors only other packages.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"As can be seen in Figure 3, the phrases in the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set include completely different relationships, which are not paraphrases.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Because of their size, the examples (Figures 2 to 4) appear at the end of the paper.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The alignment mapping is j ! i = aj from source position j to target position i = aj . The use of this alignment model raises major problems if a source word has to be aligned to several target words, e.g. when translating German compound nouns.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"input: source string f1:::fj :::fJ initialization for each cardinality c = 1; 2; ; J do for each pair (C; j), where j 2 C and jCj = c do for each target word e 2 E Qe0 (e; C; j) = p(fj je) max \u00c3\u0086;e00 j02Cnfjg fp(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j0)g words fj in the input string of length J. For the final translation each source position is considered exactly once.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Our empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We must adjoin all trees in an auxiliary tree set together as a single step in the derivation.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In our work, we demonstrate that using a simple na\u00c2\u00a8\u00c4\u00b1veBayes approach also yields substantial performance gains, without the associated training complexity.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Asking the annotator to also formulate the question is a way of arriving at more reproducible decisions.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"It may seem surprising to some readers that the interhuman agreement scores reported here are so low.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"As will be obvious later, their derivation tree sets will be local sets as are those of CFG's.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"(2009) also report results on English, but on the reduced 17 tag set, which is not comparable to ours).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Because b is a function, no additional hypothesis splitting happens.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"However, for our purposes it is not sufficient to repre\u00c2\u00ad sent the morphological decomposition of, say, plural nouns: we also need an estimate of the cost of the resulting word.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For each domain, we created a blind test set by manually annotating 40 doc uments with anaphoric chains, which represent sets of m3 (S) = ) X \u00e2\u0088\u00a9Y =S 1 \u00e2\u0088\u0092 ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) m1 (X ) \u00e2\u0088\u0097 m2 (Y ) (1) noun phrases that are coreferent (as done for MUC6 (MUC6 Proceedings, 1995)).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"To explicitly handle the word reordering between words in source and target language, we use the concept of the so-called inverted alignments as given in (Ney et al., 2000).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"If enough parsers suggest that a particular constituent belongs in the parse, we include it.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Gather phrases using keywords Now, the keyword with the top TF\/ITF score is selected for each phrase.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"(2009) on Portuguese (Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Examples are given in Table 4.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"However, in practice, unknown word models also make the distribution improper.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"For example, in the CC-domain, 96 keywords are found which have TF\/ITF scores above a threshold; some of them are shown in Figure 3.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"They also describe an application of cotraining to classifying web pages (the to feature sets are the words on the page, and other pages pointing to the page).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"For example, the independence assumptions mean that the model fails to capture the dependence between specific and more general features (for example the fact that the feature full.-string=New_York is always seen with the features contains (New) and The baseline method tags all entities as the most frequent class type (organization). contains (York) and is never seen with a feature such as contains (Group) ).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"It is. based on the traditional character set rather than the simplified character set used in Singapore and Mainland China.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"our full model yields 39.3% average error reduction across languages when compared to the basic configuration (1TW).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"One is that smaller sets sometime have meaningless keywords, like \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d in the CC-domain, or \u00e2\u0080\u009ccompare\u00e2\u0080\u009d in the PC-domain.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Evaluation of the Segmentation as a Whole.","label":0},{"source_text":"There is no global pruning.","target_text":"In Eq.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"These models generally outperform our memory consumption but are much slower, even when cached.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Quite often, though, these directives fulfill the goal of increasing annotator agreement without in fact settling the theoretical question; i.e., the directives are clear but not always very well motivated.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The features are used to represent each example for the learning algorithm.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Our implementation permits jumping to any n-gram of any length with a single lookup; this appears to be unique among language model implementations.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The 3rd block contains the mixture baselines.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"For parsing, this is a mistake, especially in the case of interrogatives.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The complexity of the algorithm is O(E3 J2 2J), where E is the size of the target language vocabulary.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"For example, let us consider a tree set containing trees of the form shown in Figure 4a.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"2.1 Part-of-speech tags.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"of Articles No.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"In the interest of testing the robustness of these combining techniques, we added a fourth, simple nonlexicalized PCFG parser.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We run the baseline Moses system for the French-English track of the 2011 Workshop on Machine Translation,9 translating the 3003-sentence test set.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Table 3 Classes of words found by ST for the test corpus.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In this way, the method reported on here will necessarily be similar to a greedy method, though of course not identical.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Word frequencies are estimated by a re-estimation procedure that involves apply\u00c2\u00ad ing the segmentation algorithm presented here to a corpus of 20 million words,8 using 8 Our training corpus was drawn from a larger corpus of mixed-genre text consisting mostly of.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"It is formally straightforward to extend the grammar to include these names, though it does increase the likelihood of overgeneration and we are unaware of any working systems that incorporate this type of name.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"If the semantic expectations do not intersect, then we know that the case- frames extract mutually exclusive types of noun phrases.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(a) I f f fi * fi :1 }'l ij 1\u00c2\u00a7: {1M m m s h e n 3 m e 0 shi2 ho u4 wo 3 cai2 ne ng 2 ke4 fu 2 zh e4 ge 4 ku n4 w h a t ti m e I just be abl e ov er co m e thi s C L dif fic 'When will I be able to overcome this difficulty?'","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The data structure was populated with 64-bit integers sampled uniformly without replacement.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Surprisingly, the non-parametric switching technique also exhibited robust behaviour in this situation.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Our final average POS tagging accuracy of 83.4% compares very favorably to the average accuracy of Berg-Kirkpatrick et al.\u2019s monolingual unsupervised state-of-the-art model (73.0%), and considerably bridges the gap to fully supervised POS tagging performance (96.6%).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"And time is short.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Each member of a set of trees can be adjoined into distinct nodes of trees in a single elementary tree set, i.e, derivations always involve the adjunction of a derived auxiliary tree set into an elementary tree set.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In approaching this problem, a variety of different methods are conceivable, including a more or less sophisticated use of machine learning.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Quite often, though, these directives fulfill the goal of increasing annotator agreement without in fact settling the theoretical question; i.e., the directives are clear but not always very well motivated.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Clearly it is possible to write a rule that states that if an analysis Modal+ Verb is available, then that is to be preferred over Noun+ Verb: such a rule could be stated in terms of (finite-state) local grammars in the sense of Mohri (1993).","label":0},{"source_text":"There is no global pruning.","target_text":"(S; C; j); Not only the coverage set C and the positions j; j0, but also the verbgroup states S; S0 are taken into account.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Clearly the percentage of productively formed words is quite small (for this particular corpus), meaning that dictionary entries are covering most of the 15 GR is .73 or 96%..","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"This is consistent with the nature of these two settings: log-linear combination, which effectively takes the intersection of IN and OUT, does relatively better on NIST, where the domains are broader and closer together.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"This allows an unbounded amount of information about two separate paths (e.g. an encoding of their length) to be combined and used to influence the later derivation.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We tried two versions of our graph-based approach: feature after the first stage of label propagation (Eq.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"This is similar to stacking the different feature instantiations into long (sparse) vectors and computing the cosine similarity between them.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"F1 85 Berkeley 80 Stanford.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We evaluate our model on seven languages exhibiting substantial syntactic variation.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This is done using a simple PCFG which is lexemebased.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"It is difficult to motivate these days why one ministry should be exempt from cutbacks \u00e2\u0080\u0094 at the expense of the others.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Reflexive pronouns with only 1 NP in scope..","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For each language under consideration, Petrov et al. (2011) provide a mapping A from the fine-grained language specific POS tags in the foreign treebank to the universal POS tags.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Given an anaphor and candidate, BABAR checks (1) whether the semantic classes of the anaphor intersect with the semantic expectations of the caseframe that extracts the candidate, and (2) whether the semantic classes of the candidate intersect with the semantic ex pectations of the caseframe that extracts the anaphor.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"More recently, Subramanya et al. (2010) defined a graph over the cliques in an underlying structured prediction model.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Phrase tables were extracted from the IN and OUT training corpora (not the dev as was used for instance weighting models), and phrase pairs in the intersection of the IN and OUT phrase tables were used as positive examples, with two alternate definitions of negative examples: The classifier trained using the 2nd definition had higher accuracy on a development set.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The percentage scores on the axis labels represent the amount of variation in the data explained by the dimension in question.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The ATB gives several different analyses to these words to indicate different types of coordination.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Language models that contain wi must also contain prefixes wi for 1 G i G k. Therefore, when the model is queried for p(wnjwn\u22121 1 ) but the longest matching suffix is wnf , it may return state s(wn1) = wnf since no longer context will be found.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Otherwise, it is set to 0.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"We can better predict the probability of an unseen hanzi occurring in a name by computing a within-class Good-Turing estimate for each radical class.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Step 3.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"As noted in Section 4.4, disk cache state is controlled by reading the entire binary file before each test begins.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In this way, the method reported on here will necessarily be similar to a greedy method, though of course not identical.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"N, portion of examples on which both classifiers give a label rather than abstaining), and the proportion of these examples on which the two classifiers agree.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We focus on this difference between the tree sets of CFG's and IG's, and formalize the notion of dependence between paths in a tree set in Section 3.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The samples from each corpus were independently evaluated.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"html 4 www.wagsoft.com\/RSTTool assigning rhetorical relations is a process loaded with ambiguity and, possibly, subjectivity.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Email: rlls@bell-labs.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"(b) After they were released...","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Like the string languages of MCTAG's, the complexity of the path set increases as the cardinality of the elementary tee sets increases, though both the string languages and path sets will always be semilinear.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The quasi-monotone search performs best in terms of both error rates mWER and SSER.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"76 16.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"(Hartmann 1984), for example, used the term Reliefgebung to characterize the distibution of main and minor information in texts (similar to the notion of nuclearity in RST).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"A secondary reference resolution classifier has information on the class assigned by the primary classifier.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Unlabeled examples in the named-entity classification problem can reduce the need for supervision to a handful of seed rules.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"As indicated by bolding, for seven out of eight languages the improvements of the \u201cWith LP\u201d setting are statistically significant with respect to the other models, including the \u201cNo LP\u201d setting.11 Overall, it performs 10.4% better than the hitherto state-of-the-art feature-HMM baseline, and 4.6% better than direct projection, when we macro-average the accuracy over all languages.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Two lists, Corporate-Suffix-List (for corporate suffixes) and Person-Prefix-List (for person prefixes), are collected from the training data.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"7 68.3 56.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"However, if we consider precision, recall and Fmeasure on non-projective dependencies only, as shown in Table 6, some differences begin to emerge.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Therefore, performance is more closely tied to the underlying data structure than to the cache.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Conditioned on T , features of word types W are drawn.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"For 2 < n < N, we use a hash table mapping from the n-gram to the probability and backoff3.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In this specific case, as these two titles could fill the same column of an IE table, we regarded them as paraphrases for the evaluation.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"However, when grammatical relations like subject and object are evaluated, parsing performance drops considerably (Green et al., 2009).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Using a wide-coverage morphological analyzer based on (Itai et al., 2006) should cater for a better coverage, and incorporating lexical probabilities learned from a big (unannotated) corpus (cf.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Finally, we thank Kuzman Ganchev and the three anonymous reviewers for helpful suggestions and comments on earlier drafts of this paper.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We adapt the string pumping lemma for the class of languages corresponding to the complexity of the path set.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For a sequence of hanzi that is a possible name, we wish to assign a probability to that sequence qua name.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The probability of the classes assigned to the words in a sentence in a document is defined as follows: where is determined by the maximum entropy classifier.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The key to the methods we describe is redundancy in the unlabeled data.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"A straightforward way to find the shortest tour is by trying all possible permutations of the n cities.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"A total of 13,976 phrases were grouped.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"NN \u00ef\u00bf\u00bd .e NP NNP NP DTNNP NN \u00ef\u00bf\u00bd .e NP NP NNP NP Table 5: Evaluation of 100 randomly sampled variation nuclei types.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We evaluate our model on seven languages exhibiting substantial syntactic variation.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Previous work on morphological and syntactic disambiguation in Hebrew used different sets of data, different splits, differing annotation schemes, and different evaluation measures.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(1991}, Gu and Mao (1994), and Nie, Jin, and Hannan (1994).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Applications such as machine translation use language model probability as a feature to assist in choosing between hypotheses.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"17 They also provide a set of title-driven rules to identify names when they occur before titles such as $t. 1: xianlshengl 'Mr.' or i:l:itr!J tai2bei3 shi4zhang3 'Taipei Mayor.'","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The type-level posterior term can be computed according to, P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2) \u00e2\u0088\u009d Note that each round of sampling Ti variables takes time proportional to the size of the corpus, as with the standard token-level HMM.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"2.1 Part-of-speech tags.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Furthermore, the systematic way in which particles are prefixed to one another and onto an open-class category gives rise to a distinct sort of morphological ambiguity: space-delimited tokens may be ambiguous between several different segmentation possibilities.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The Wang, Li, and Chang system fails on fragment (b) because their system lacks the word youlyoul 'soberly' and misinterpreted the thus isolated first youl as being the final hanzi of the preceding name; similarly our system failed in fragment (h) since it is missing the abbreviation i:lJI!","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"As shown in Table 3, the proportion of sentences containing some non-projective dependency ranges from about 15% in DDT to almost 25% in PDT.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The following rule corresponds to the above derivation, where 71, , 7k are derived from the auxiliary trees , , fik, respectively. for all addresses n in some elementary tree at which 7' can be adjoined.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each co-occurrence relation (noun\/caseframe for CFLex, and caseframe\/caseframe for CFNet), BABAR computes its log-likelihood value and looks it up in the \u00cf\u00872 table to obtain a confidence level.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The cost is computed as follows, where N is the corpus size and f is the frequency: (1) Besides actual words from the base dictionary, the lexicon contains all hanzi in the Big 5 Chinese code\/ with their pronunciation(s), plus entries for other characters that can be found in Chinese text, such as Roman letters, numerals, and special symbols.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Almost all annotators expressed their preference to move to a ranking-based evaluation in the future.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The treebank has two versions, v1.0 and v2.0, containing 5001 and 6501 sentences respectively.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"For all these annotation taks, G\u00c2\u00a8otze developed a series of questions (essentially a decision tree) designed to lead the annotator to the ap propriate judgement.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We compare three hash tables: our probing implementation, GCC\u2019s hash set, and Boost\u2019s8 unordered.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We use a patched version of BitPar allowing for direct input of probabilities instead of counts.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"While Berg-Kirkpatrick et al.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"These results are promising and there are several avenues for improving on these results.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"When a company buys another company, a paying event can occur, but these two phrases do not indicate the same event.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We further thank Khalil Simaan (ILLCUvA) for his careful advise concerning the formal details of the proposal.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"As can be seen in Table 4, our training data is a lot less than those used by MENE and IdentiFinder3.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"All annotations are done with specific tools and in XML; each layer has its own DTD.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thus, TAG's can not give analyses in which dependencies between arbitrarily large branches exist.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"For the \u00e2\u0080\u0098core\u00e2\u0080\u0099 portion of PCC, we found that on average, 35% of the coherence relations in our RST annotations are explicitly signalled by a lexical connective.6 When adding the fact that connectives are often ambiguous, one has to conclude that prospects for an automatic analysis of rhetorical structure using shallow methods (i.e., relying largely on connectives) are not bright \u00e2\u0080\u0094 but see Sections 3.2 and 3.3 below.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"DempsterShafer handles this by re-normalizing all the belief values with respect to only the non-null sets (this is the purpose of the denominator in Equation 1).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The anaphor and antecedent appear in boldface.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Note that Wang, Li, and Chang's.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We call these N \u2212 1 words state.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We will refer to our system as MENERGI (Maximum Entropy Named Entity Recognizer using Global Information).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Brin 98) ,describes a system for extracting (author, book-title) pairs from the World Wide Web using an approach that bootstraps from an initial seed set of examples.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Either save money at any cost - or give priority to education.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"For alif with hamza, normalization can be seen as another level of devocalization.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"1 61.2 43.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"IL+-1Proof: Assume a pair of crossing constituents appears in the output of the constituent voting technique using k parsers.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Until now, all evaluations of Arabic parsing\u00e2\u0080\u0094including the experiments in the previous section\u00e2\u0080\u0094have assumed gold segmentation.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The BerkeleyLM direct-mapped cache is in principle faster than caches implemented by RandLM and by IRSTLM, so we may write a C++ equivalent implementation as future work.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"To approximate these baselines, we implemented a very simple sentence selection algorithm in which parallel sentence pairs from OUT are ranked by the perplexity of their target half according to the IN language model.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We have presented a method for unsupervised part- of-speech tagging that considers a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(4) gives In order to minimize Zt, at each iteration the final algorithm should choose the weak hypothesis (i.e., a feature xt) which has values for W+ and W_ that minimize Equ.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In German, the verbgroup usually consists of a left and a right verbal brace, whereas in English the words of the verbgroup usually form a sequence of consecutive words.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"This is equivalent to the assumption used in probability estimation for na\u00efve Bayes classifiers, namely that the attribute values are conditionally independent when the target value is given.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"German English Training: Sentences 58 073 Words 519 523 549 921 Words* 418 979 453 632 Vocabulary Size 7939 4648 Singletons 3454 1699 Test-147: Sentences 147 Words 1 968 2 173 Perplexity { 26:5 Table 4: Multi-reference word error rate (mWER) and subjective sentence error rate (SSER) for three different search procedures.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Note that because we extracted only high-confidence alignments, many foreign vertices will not be connected to any English vertices.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Section 3 describes the complete coreference resolution model, which uses the contextual role knowledge as well as more traditional coreference features.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Finally, the focus\/background partition is annotated, together with the focus question that elicits the corresponding answer.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"These enable much larger models in memory, compensating for lost accuracy.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Word frequencies are estimated by a re-estimation procedure that involves apply\u00c2\u00ad ing the segmentation algorithm presented here to a corpus of 20 million words,8 using 8 Our training corpus was drawn from a larger corpus of mixed-genre text consisting mostly of.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"They also describe an application of cotraining to classifying web pages (the to feature sets are the words on the page, and other pages pointing to the page).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The 2nd block contains the IR system, which was tuned by selecting text in multiples of the size of the EMEA training corpus, according to dev set performance.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Statistics for all data sets are shown in Table 2.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This work was supported in part by the National Science Foundation under grant IRI9704240.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This style of naming is never required and seems to be losing currency.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"\u2014 I would also like to point out to commissioner Liikanen that it is not easy to take a matter to a national court.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"This limitation is the obstacle to making the technology \u00e2\u0080\u009copen domain\u00e2\u0080\u009d.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"One conclusion drawn from this annotation effort was that for humans and machines alike, 2 www.sfs.nphil.unituebingen.de\/Elwis\/stts\/ stts.html 3 www.coli.unisb.de\/sfb378\/negra-corpus\/annotate.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Note that hanzi that are not grouped into dictionary words (and are not identified as single\u00c2\u00ad hanzi words), or into one of the other categories of words discussed in this paper, are left unattached and tagged as unknown words.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The search starts in the hypothesis (I; f;g; 0).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"To check for this, we do pairwise bootstrap resampling: Again, we repeatedly sample sets of sentences, this time from both systems, and compare their BLEU scores on these sets.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The importance of this property becomes clear in contrasting theories underlying GPSG (Gazdar, Klein, Pulluna, and Sag, 1985), and GB (as described by Berwick, 1984) with those underlying LFG and FUG.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In our coreference resolver, we define \u00ce\u00b8 to be the set of all candidate antecedents for an anaphor.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The sentence length probability p(JjI) is omitted without any loss in performance.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"A related point is that mutual information is helpful in augmenting existing electronic dictionaries, (cf.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Giving a recognition algorithm for LCFRL's involves describing the substrings of the input that are spanned by the structures derived by the LCFRS's and how the composition operation combines these substrings.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"It is difficult to directly compare the Matsoukas et al results with ours, since our out-of-domain corpus is homogeneous; given heterogeneous training data, however, it would be trivial to include Matsoukas-style identity features in our instance-weighting model.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Manual and Automatic Evaluation of Machine Translation between European Languages","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Note that Wang, Li, and Chang's.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In particular, it may not be possible to learn functions fi (x f2(x2,t) for i = m + 1...n: either because there is some noise in the data, or because it is just not realistic to expect to learn perfect classifiers given the features used for representation.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":", December, then the feature MonthName is set to 1.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Within this framework, we use features intended to capture degree of generality, including the output from an SVM classifier that uses the intersection between IN and OUT as positive examples.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Recently, combination techniques have been investigated for part of speech tagging with positive results (van Halteren et al., 1998; Brill and Wu, 1998).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"It's not clear how to apply these methods in the unsupervised case, as they required cross-validation techniques: for this reason we use the simpler smoothing method shown here. input to the unsupervised algorithm is an initial, &quot;seed&quot; set of rules.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Finally, this effort is part of a much larger program that we are undertaking to develop stochastic finite-state methods for text analysis with applications to TIS and other areas; in the final section of this paper we will briefly discuss this larger program so as to situate the work discussed here in a broader context.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We also performed experiments to evaluate the impact of each type of contextual role knowledge separately.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"One implementation issue deserves some elaboration.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Both BBN and NYU have tagged their own data to supplement the official training data.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"At each iteration the algorithm increases the number of rules, while maintaining a high level of agreement between the spelling and contextual decision lists.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For terrorism, BABAR generated 5,078 resolutions: 2,386 from lexical seeding and 2,692 from syntactic seeding.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In future work, we plan to follow-up on this approach and investigate other ways that contextual role knowledge can be used.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"11 taTweel (-) is an elongation character used in Arabic script to justify text.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Annotation consistency is important in any supervised learning task.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Second, rather than relying on a division of the corpus into manually-assigned portions, we use features intended to capture the usefulness of each phrase pair.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The government has to make a decision, and do it quickly.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We introduce several new ideas.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We hope that this will allow practitioners to apply our approach directly to languages for which no resources are available.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Parent Head Modif er Dir # gold F1 Label # gold F1 NP NP TAG R 946 0.54 ADJP 1216 59.45 S S S R 708 0.57 SBAR 2918 69.81 NP NP ADJ P R 803 0.64 FRAG 254 72.87 NP NP N P R 2907 0.66 VP 5507 78.83 NP NP SBA R R 1035 0.67 S 6579 78.91 NP NP P P R 2713 0.67 PP 7516 80.93 VP TAG P P R 3230 0.80 NP 34025 84.95 NP NP TAG L 805 0.85 ADVP 1093 90.64 VP TAG SBA R R 772 0.86 WHN P 787 96.00 S VP N P L 961 0.87 (a) Major phrasal categories (b) Major POS categories (c) Ten lowest scoring (Collins, 2003)-style dependencies occurring more than 700 times Table 8: Per category performance of the Berkeley parser on sentence lengths \u00e2\u0089\u00a4 70 (dev set, gold segmentation).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"9 61.0 44.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"5.2 Setup.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Our original hope in combining these parsers is that their errors are independently distributed.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Te rro ris m Na tur al Dis ast ers mu rde r of < NP > kill ed <p atie nt > <a ge nt > da ma ged wa s inj ure d in < NP > <a ge nt > rep ort ed <a ge nt > add ed <a ge nt > occ urr ed cau se of < NP > <a ge nt > stat ed <a ge nt > add ed <a ge nt > wr eak ed <a ge nt > cro sse d per pet rat ed <p atie nt > con de mn ed <p atie nt > dri ver of < NP > <a ge nt > car ryi ng Figure 1: Caseframe Network Examples Figure 1 shows examples of caseframes that co-occur in resolutions, both in the terrorism and natural disaster domains.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We evaluate the time and memory consumption of each data structure by computing perplexity on 4 billion tokens from the English Gigaword corpus (Parker et al., 2009).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Moreover, if we want to test on a huge test corpus, indexing the whole corpus might prove computationally expensive.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This method, one instance of which we term the \"greedy algorithm\" in our evaluation of our own system in Section 5, involves starting at the beginning (or end) of the sentence, finding the longest word starting (ending) at that point, and then repeating the process starting at the next (previous) hanzi until the end (begin\u00c2\u00ad ning) of the sentence is reached.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"This is the first time that we organized a large-scale manual evaluation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Obviously, the presence of a title after a potential name N increases the probability that N is in fact a name.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The sources of our dictionaries are listed in Table 2.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"This fully generative model caters for real interaction between the syntactic and morphological levels as a part of a single coherent process.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The pseudo-code describing the algorithm is given in Fig.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"This fact annoyed especially his dog...).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Yet, some hanzi are far more probable in women's names than they are in men's names, and there is a similar list of male-oriented hanzi: mixing hanzi from these two lists is generally less likely than would be predicted by the independence model.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Since the inclusion of out-ofdomain test data was a very late decision, the participants were not informed of this.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The contextual role knowledge had the greatest impact on pronouns: +13% recall for terrorism and +15% recall for disasters, with a +1% precision gain in terrorism and a small precision drop of -3% in disasters.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"It has been shown for English (Wang and Hirschberg 1992; Hirschberg 1993; Sproat 1994, inter alia) that grammatical part of speech provides useful information for these tasks.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The corresponding token words w are drawn conditioned on t and \u00ce\u00b8.2 Our full generative model is given by: K P (\u00cf\u0086, \u00ce\u00b8|T , \u00ce\u00b1, \u00ce\u00b2) = n (P (\u00cf\u0086t|\u00ce\u00b1)P (\u00ce\u00b8t|T , \u00ce\u00b1)) t=1 The transition distribution \u00cf\u0086t for each tag t is drawn according to DIRICHLET(\u00ce\u00b1, K ), where \u00ce\u00b1 is the shared transition and emission distribution hyperparameter.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Proper-Name Identification.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Proper names that match are resolved with each other.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For example, hanzi containing the INSECT radical !R tend to denote insects and other crawling animals; examples include tr wal 'frog,' feng1 'wasp,' and !Itt she2 'snake.'","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Time for Moses itself to load, including loading the language model and phrase table, is included.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"58 95.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In recent years, coreference resolvers have been evaluated as part of MUC6 and MUC7 (MUC7 Proceedings, 1998).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Since the segmentation corresponds to the sequence of words that has the lowest summed unigram cost, the segmenter under discussion here is a zeroth-order model.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We introduce several new ideas.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Other approaches encode sparsity as a soft constraint.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"27 80.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"92 76.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The experimental tests are carried out on the Verbmobil task (GermanEnglish, 8000-word vocabulary), which is a limited-domain spoken-language task.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"This can be seen as a rough approximation of Yarowsky and Ngai (2001).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"3.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Since we could not bias the subjects towards a particular segmentation and did not presume linguistic sophistication on their part, the instructions were simple: subjects were to mark all places they might plausibly pause if they were reading the text aloud.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In this section we present a partial evaluation of the current system, in three parts.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Syntactic decoders, such as cdec (Dyer et al., 2010), build state from null context then store it in the hypergraph node for later extension.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In this paper, Section 2 begins by explaining how contextual role knowledge is represented and learned.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"To approximate these baselines, we implemented a very simple sentence selection algorithm in which parallel sentence pairs from OUT are ranked by the perplexity of their target half according to the IN language model.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Each knowledge source then assigns a probability estimate to each candidate, which represents its belief that the candidate is the antecedent for the anaphor.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"2.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"This measure has the advantage of being completely automatic.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Each trie node contains a sorted array of entries and they use binary search.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"For the error counts, a range from 0:0 to 1:0 is used.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The set of candidate constituents comes from the union of all the constituents suggested by the member parsers.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The DempsterShafer rule for combining pdfs is: to {C}, meaning that it is 70% sure the correct hypothesis is C. The intersection of these sets is the null set because these beliefs are contradictory.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The results in Table 2 were achieved on the development set.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"splitPUNC restores the convention of the WSJ.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"7 Conclusion and Future Work.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Therefore, for n-gram wn1 , all leftward extensions wn0 are an adjacent block in the n + 1-gram array.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"The links can solve the problem.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Given an anaphor and candidate, BABAR checks (1) whether the semantic classes of the anaphor intersect with the semantic expectations of the caseframe that extracts the candidate, and (2) whether the semantic classes of the candidate intersect with the semantic ex pectations of the caseframe that extracts the anaphor.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Two of the Mainlanders also cluster close together but, interestingly, not particularly close to the Taiwan speakers; the third Mainlander is much more similar to the Taiwan speakers.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"For the experiments in this paper we made a couple of additional modifications to the CoBoost algorithm.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"During the search process, a partial hypothesis is extended by choosing a source sentence position, which has not been aligned with a target sentence position yet.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE has only been tested on MUC7.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J) and the lexicon probability p(fbi jei) are obtained by relative frequency estimates from the Viterbi alignment path after the final training iteration.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Since the inclusion of out-ofdomain test data was a very late decision, the participants were not informed of this.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"3 The Coreference Resolution Model.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Features and context were initially introduced into the models, but they refused to offer any gains in performance.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The accuracy results for segmentation, tagging and parsing using our different models and our standard data split are summarized in Table 1.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Note that we check for , the word preceding the consecutive sequence of initCaps tokens, since person prefixes like Mr., Dr., etc are not part of person names, whereas corporate suffixes like Corp., Inc., etc are part of corporate names.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"(2009b) evaluated the Bikel parser using the same ATB split, but only reported dev set results with gold POS tags for sentences of length \u00e2\u0089\u00a4 40.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We weight edges using a unigram language model estimated with Good- Turing smoothing.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"information structure.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"However, in practice, unknown word models also make the distribution improper.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"It should be clear from the onset that the particle b (\u201cin\u201d) in \u2018bcl\u2019 may then attach higher than the bare noun cl (\u201cshadow\u201d).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Lexicon Feature: The string of the token is used as a feature.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We used this data to build an unpruned ARPA file with IRSTLM\u2019s improved-kneser-ney option and the default three pieces.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For locations, tokens are matched against unigrams, and sequences of two consecutive tokens are matched against bigrams.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"However, there are phrases which express the same meanings even though they do not share the same keyword.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Searching a probing hash table consists of hashing the key, indexing the corresponding bucket, and scanning buckets until a matching key is found or an empty bucket is encountered, in which case the key does not exist in the table.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Figure 1 depicts the lattice for a 2-words sentence bclm hneim.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The less favored reading may be selected in certain contexts, however; in the case of , for example, the nominal reading jiang4 will be selected if there is morphological information, such as a following plural affix ir, menD that renders the nominal reading likely, as we shall see in Section 4.3.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The 4th block contains instance-weighting models trained on all features, used within a MAP TM combination, and with a linear LM mixture.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"3 68.9 50.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Each pattern represents the role that a noun phrase plays in the surrounding context.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"3 54.4 33.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"On the first of these-the B set-our system had 64% recall and 86% precision; on the second-the C set-it had 33% recall and 19% precision.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The new algorithm, which we call CoBoost, uses labeled and unlabeled data and builds two classifiers in parallel.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"conceptual relationship in the discourse.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"On the one hand, the type-level error rate is not calibrated for the number of n-grams in the sample.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Evalb is a Java re-implementation of the standard labeled precision\/recall metric.12 The ATB gives all punctuation a single tag.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We use the universal POS tagset of Petrov et al. (2011) in our experiments.10 This set C consists of the following 12 coarse-grained tags: NOUN (nouns), VERB (verbs), ADJ (adjectives), ADV (adverbs), PRON (pronouns), DET (determiners), ADP (prepositions or postpositions), NUM (numerals), CONJ (conjunctions), PRT (particles), PUNC (punctuation marks) and X (a catch-all for other categories such as abbreviations or foreign words).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In addition, each feature function is a binary function.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"\u00e2\u0080\u00a2 Bridging links: the annotator is asked to specify the type as part-whole, cause-effect (e.g., She had an accident.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We can make several observations on the cause of errors.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"16 As one reviewer points out, one problem with the unigram model chosen here is that there is still a. tendency to pick a segmentation containing fewer words.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Search t0 CPU time #search mWER Method [sec] error [%] QmS 0.0 0.07 108 42:6 1.0 0.13 85 37:8 2.5 0.35 44 36:6 5.0 1.92 4 34:6 10.0 10.6 0 34:5 IbmS 0.0 0.14 108 43:4 1.0 0.3 84 39:5 2.5 0.8 45 39:1 5.0 4.99 7 38:3 10.0 28.52 0 38:2 Table 6 shows example translations obtained by the three different approaches.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In the named entity domain these rules were Each of these rules was given a strength of 0.9999.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In this paper, we outline how such family of formalisms can be defined, and show that like CFG's, each member possesses a number of desirable linguistic and computational properties: in particular, the constant growth property and polynomial recognizability.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"It can be easily shown from Thatcher's result that the path set of every local set is a regular set.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In this work we extended the AdaBoost.MH (Schapire and Singer 98) algorithm to the cotraining case.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"On average, 6 reference translations per automatic translation are available.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We performed three experiments to evaluate our techniques.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The English data comes from the WSJ portion of the Penn Treebank and the other languages from the training set of the CoNLL-X multilingual dependency parsing shared task.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, a story can mention \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d, \u00e2\u0080\u009cthe White House\u00e2\u0080\u009d, or \u00e2\u0080\u009cthe weather\u00e2\u0080\u009d without any prior referent in the story.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The government has to make a decision, and do it quickly.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Assigning rhetorical relations thus poses questions that can often be answered only subjectively.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For example, the Wang, Li, and Chang system fails on the sequence 1:f:p:]nian2 nei4 sa3 in (k) since 1F nian2 is a possible, but rare, family name, which also happens to be written the same as the very common word meaning 'year.'","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We consider the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","label":1},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"If \u00e2\u0080\u009cgun\u00e2\u0080\u009d and \u00e2\u0080\u009crevolver\u00e2\u0080\u009d refer to the same object, then it should also be acceptable to say that Fred was \u00e2\u0080\u009ckilled with a gun\u00e2\u0080\u009d and that the burglar \u00e2\u0080\u009cfireda revolver\u00e2\u0080\u009d.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Having found (spelling, context) pairs in the parsed data, a number of features are extracted.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Much could be done to further reduce memory consumption.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In addition, this formulation results in a dramatic reduction in the number of model parameters thereby, enabling unusually rapid training.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For previously unseen hanzi in given names, Chang et al. assign a uniform small cost; but we know that some unseen hanzi are merely acci\u00c2\u00ad dentally missing, whereas others are missing for a reason-for example, because they have a bad connotation.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The resulting parsers surpass the best previously published performance results for the Penn Treebank.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Specifically, we assume each word type W consists of feature-value pairs (f, v).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"For each trigram type x2 x3 x4 in a sequence x1 x2 x3 x4 x5, we count how many times that trigram type co-occurs with the different instantiations of each concept, and compute the point-wise mutual information (PMI) between the two.5 The similarity between two trigram types is given by summing over the PMI values over feature instantiations that they have in common.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"3 68.4 49.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Obviously, the presence of a title after a potential name N increases the probability that N is in fact a name.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Person\u00e2\u0080\u009d domain, 618 keywords are found, and in the \u00e2\u0080\u009cCountry \u00e2\u0080\u0093 Country\u00e2\u0080\u009d domain, 303 keywords are found.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Annotation of syntactic structure for the core corpus has just begun.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"In our case multi-threading is trivial because our data structures are read-only and uncached.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"0750271 and by the DARPA GALE program.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We picked two domains, the CC-domain and the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain (PC-domain), for the evaluation, as the entire system output was too large to evaluate.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"2.1 Inverted Alignments.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"At most one feature in this group will be set to 1.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"mein 5.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"An anti-greedy algorithm, AG: instead of the longest match, take the.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"This limitation is the obstacle to making the technology \u00e2\u0080\u009copen domain\u00e2\u0080\u009d.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"These performance gains transfer to improved system runtime performance; though we focused on Moses, our code is the best lossless option with cdec and Joshua.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(b) F.i'JJI!","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(Hearst 92) describes a method for extracting hyponyms from a corpus (pairs of words in &quot;isa&quot; relations).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We would like to thank Ryan McDonald for numerous discussions on this topic.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We focus on phrases which connect two Named Entities (NEs), and proceed in two stages.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This suggests that the backoff model is as reasonable a model as we can use in the absence of further information about the expected cost of a plural form.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The estimation of the probabilities in the model is carried out as shown in Equation 4.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Future work along these lines will incorporate other layers of annotation, in particular the syntax information.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Thus, most broad-coverage parsers based on dependency grammar have been restricted to projective structures.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For the disasters domain, 8245 texts were used for training and the 40 test documents contained 447 anaphoric links.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"2 We now introduce a new algorithm for learning from unlabeled examples, which we will call DLCoTrain (DL stands for decision list, the term Cotrain is taken from (Blum and Mitchell 98)).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"It can also be seen clearly in this plot that two of the Taiwan speakers cluster very closely together, and the third Tai\u00c2\u00ad wan speaker is also close in the most significant dimension (the x axis).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We will also directly compare with a baseline similar to the Matsoukas et al approach in order to measure the benefit from weighting phrase pairs (or ngrams) rather than full sentences.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"A Single Generative Model for Joint Morphological Segmentation and Syntactic Parsing","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"An interesting observation is that while vertical markovization benefits all our models, its effect is less evident in Cohen and Smith.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"besuchen 9.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The horizontal axis in this plot represents the most significant dimension, which explains 62% of the variation.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We paid particular attention to minimize the number of free parameters, and used the same hyperparameters for all language pairs, rather than attempting language-specific tuning.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"LCFRS's have only been loosely defined in this paper; we have yet to provide a complete set of formal properties associated with members of this class.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"7 Unlike Dickinson (2005), we strip traces and only con-.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"They are also labelled for their topicality (yes \/ no), and this annotation is accompanied by a confidence value assigned by the annotator (since it is a more subjective matter).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"The linear LM (lin lm), TM (lin tm) and MAP TM (map tm) used with non-adapted counterparts perform in all cases slightly worse than the log-linear combination, which adapts both LM and TM components.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Three state-of-the-art statistical parsers are combined to produce more accurate parses, as well as new bounds on achievable Treebank parsing accuracy.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Upon identifying an anaphoric expression (currently restricted to: pronouns, prepositional adverbs, definite noun phrases), the an- notator first marks the antecedent expression (currently restricted to: various kinds of noun phrases, prepositional phrases, verb phrases, sentences) and then establishes the link between the two.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"f1; ; Jg denotes a coverage set including all positions from the starting position 1 to position J and j 2 fJ \u00f4\u0080\u0080\u0080L; ; Jg.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"In our experiments, we used the same set of features as BergKirkpatrick et al. (2010): an indicator feature based In a traditional Markov model, the emission distribution P\u0398(Xi = xi  |Zi = zi) is a set of multinomials.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"MITLM 0.4 (Hsu and Glass, 2008) is mostly designed for accurate model estimation, but can also compute perplexity.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The possible analyses of a surface token pose constraints on the analyses of specific segments.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In this section we present a partial evaluation of the current system, in three parts.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"On the MUC6 data, Bikel et al.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"rhetorical analysis We are experimenting with a hybrid statistical and knowledge-based system for discourse parsing and summarization (Stede 2003), (Hanneforth et al. 2003), again targeting the genre of commentaries.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In recent years, coreference resolvers have been evaluated as part of MUC6 and MUC7 (MUC7 Proceedings, 1998).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"A more forceful approach for encoding sparsity is posterior regularization, which constrains the posterior to have a small number of expected tag assignments (Grac\u00c2\u00b8a et al., 2009).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Arguably this consists of about three phonological words.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"4 To be sure, it is not always true that a hanzi represents a syllable or that it represents a morpheme.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"), and thosethat begin with a verb (\u00ef\u00bf\u00bd ub..i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"BABAR uses the log-likelihood statistic (Dunning, 1993) to evaluate the strength of a co-occurrence relationship.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For example, in the sentence that starts with \u00e2\u0080\u009cBush put a freeze on . . .","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"3.1 Lexicon Component.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"5 64.7 42.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Our results show that BABAR achieves good performance in both domains, and that the contextual role knowledge improves performance, especially on pronouns.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"(1998) did make use of information from the whole document.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We identified three ways that contextual roles can be exploited: (1) by identifying caseframes that co-occur in resolutions, (2) by identifying nouns that co-occur with case- frames and using them to crosscheck anaphor\/candidate compatibility, (3) by identifying semantic classes that co- occur with caseframes and using them to crosscheck anaphor\/candidate compatability.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"We hope that this will allow practitioners to apply our approach directly to languages for which no resources are available.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"In (Reitter, Stede 2003) we went a different way and suggested URML5, an XML format for underspecifying rhetorical structure: a number of relations can be assigned instead of a single one, competing analyses can be represented with shared forests.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"SSER: subjective sentence error rate: For a more detailed analysis, the translations are judged by a human test person.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"5 60.6 Table 3: Multilingual Results: We report token-level one-to-one and many-to-one accuracy on a variety of languages under several experimental settings (Section 5).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The feature-HMM model works better for all languages, generalizing the results achieved for English by Berg-Kirkpatrick et al. (2010).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The TRIE model continues to use the least memory of ing (-P) with MAP POPULATE, the default.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Searching a probing hash table consists of hashing the key, indexing the corresponding bucket, and scanning buckets until a matching key is found or an empty bucket is encountered, in which case the key does not exist in the table.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We use a patched version of BitPar allowing for direct input of probabilities instead of counts.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"This offers the well-known advantages for inter- changability, but it raises the question of how to query the corpus across levels of annotation.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Fourth, we show how to build better models for three different parsers.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Initially, we planned to compare the semantic classes of an anaphor and a candidate and infer that they might be coreferent if their semantic classes intersected.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"As a result, Arabic sentences are usually long relative to English, especially after Length English (WSJ) Arabic (ATB) \u00e2\u0089\u00a4 20 41.9% 33.7% \u00e2\u0089\u00a4 40 92.4% 73.2% \u00e2\u0089\u00a4 63 99.7% 92.6% \u00e2\u0089\u00a4 70 99.9% 94.9% Table 2: Frequency distribution for sentence lengths in the WSJ (sections 2\u00e2\u0080\u009323) and the ATB (p1\u00e2\u0080\u00933).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The basic word order is VSO, but SVO, VOS, and VO configurations are also possible.2 Nouns and verbs are created by selecting a consonantal root (usually triliteral or quadriliteral), which bears the semantic core, and adding affixes and diacritics.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We obtained positive results using a very simple phrase-based system in two different adaptation settings: using English\/French Europarl to improve a performance on a small, specialized medical domain; and using non-news portions of the NIST09 training material to improve performance on the news-related corpora.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Supervised methods have been applied quite successfully to the full MUC named-entity task (Bikel et al. 97).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Particular relations are also consistent with particular hypotheses about the segmentation of a given sentence, and the scores for particular relations can be incremented or decremented depending upon whether the segmentations with which they are consistent are \"popular\" or not.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Our primary goal is to exploit the resources that are most appropriate for the task at hand, and our secondary goal is to allow for comparison of our models\u2019 performance against previously reported results.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"One key issue here is to seek a discourse-based model of information structure.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Table 4 shows the results.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Twentieth-century linguistic work on Chinese (Chao 1968; Li and Thompson 1981; Tang 1988,1989, inter alia) has revealed the incorrectness of this traditional view.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Since the parameter and token components will remain fixed throughout experiments, we briefly describe each.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"RandLM is the clear winner in RAM utilization, but is also slower and lower quality.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"4 7 . 3 8 . 9 2 8 . 8 2 0 . 7 3 2 . 3 3 5 . 2 2 9 . 6 2 7 . 6 1 4 . 2 4 2 . 8 4 5 . 9 4 4 . 3 6 0 . 6 6 1 . 5 4 9 . 9 3 3 . 9 Table 6: Type-level Results: Each cell report the type- level accuracy computed against the most frequent tag of each word type.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"lla\/llb and 14a\/14b respectively).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"However, our full model takes advantage of word features not present in Grac\u00c2\u00b8a et al.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The method being described-henceforth ST..","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"In this paper we have proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In this paper, we offer broad insight into the underperformance of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Fall.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For example, the Wang, Li, and Chang system fails on the sequence 1:f:p:]nian2 nei4 sa3 in (k) since 1F nian2 is a possible, but rare, family name, which also happens to be written the same as the very common word meaning 'year.'","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The weak learner for two-class problems computes a weak hypothesis h from the input space into the reals (h : 2x -4 R), where the sign4 of h(x) is interpreted as the predicted label and the magnitude I h(x)I is the confidence in the prediction: large numbers for I h(x)I indicate high confidence in the prediction, and numbers close to zero indicate low confidence.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Thus, the effects of spontaneous speech are present in the corpus, e.g. the syntactic structure of the sentence is rather less restricted, however the effect of speech recognition errors is not covered.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Note that we check for , the word preceding the consecutive sequence of initCaps tokens, since person prefixes like Mr., Dr., etc are not part of person names, whereas corporate suffixes like Corp., Inc., etc are part of corporate names.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"To this end, we picked 100 sentences at random containing 4,372 total hanzi from a test corpus.14 (There were 487 marks of punctuation in the test sentences, including the sentence-final periods, meaning that the average inter-punctuation distance was about 9 hanzi.)","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Search t0 CPU time #search mWER Method [sec] error [%] QmS 0.0 0.07 108 42:6 1.0 0.13 85 37:8 2.5 0.35 44 36:6 5.0 1.92 4 34:6 10.0 10.6 0 34:5 IbmS 0.0 0.14 108 43:4 1.0 0.3 84 39:5 2.5 0.8 45 39:1 5.0 4.99 7 38:3 10.0 28.52 0 38:2 Table 6 shows example translations obtained by the three different approaches.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Table 5 provides insight into the behavior of different models in terms of the tagging lexicon they generate.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Still, from a theoretical point of view, projective parsing of non-projective structures has the drawback that it rules out perfect accuracy even as an asymptotic goal.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This is not to say that a set of standards by which a particular segmentation would count as correct and another incorrect could not be devised; indeed, such standards have been proposed and include the published PRCNSC (1994) and ROCLING (1993), as well as the unpublished Linguistic Data Consortium standards (ca.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Applications The discovered paraphrases have multiple applications.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Like CFG's, the choice is predetermined by a finite number of rules encapsulated in the grammar.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"At the phrasal level, we remove all function tags and traces.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"vierten 12.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The structure uses linear probing hash tables and is designed for speed.","label":1},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Figure 2 shows an excerpt of a sentence from the Italian test set and the tags assigned by four different models, as well as the gold tags.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Within this framework, we use features intended to capture degree of generality, including the output from an SVM classifier that uses the intersection between IN and OUT as positive examples.","label":0},{"source_text":"Here we present two algorithms.","target_text":"To prevent this we &quot;smooth&quot; the confidence by adding a small value, e, to both W+ and W_, giving at = Plugging the value of at from Equ.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Thus, as in CFG's, at any point in the derivation, the set of structures that can be applied is determined only by a finite set of rules encapsulated by the grammar.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"These enable much larger models in memory, compensating for lost accuracy.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In this paper, we show that the maximum entropy framework is able to make use of global information directly, and achieves performance that is comparable to the best previous machine learning-based NERs on MUC6 and MUC7 test data.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The developers aimed to reduce memory consumption at the expense of time.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"handled given appropriate models.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"3.1 Word ReOrdering with Verbgroup.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Finally, note that while most feature concepts are lexicalized, others, such as the suffix concept, are not.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The inventions disclosed herein are the subject of a patent application owned by the University of Utah and licensed on an exclusive basis to Attensity Corporation.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(1992).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Examining the word fidanzato for the \u201cNo LP\u201d and \u201cWith LP\u201d models is particularly instructive.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"(Thus the domain of the dev and test corpora matches IN.)","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"computing the precision of the other's judgments relative to this standard.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A related point is that mutual information is helpful in augmenting existing electronic dictionaries, (cf.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"A promising direction for future work is to explicitly model a distribution over tags for each word type.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For example, if is found in the list of person first names, the feature PersonFirstName is set to 1.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We have studied the structural descriptions (tree sets) that can be assigned by various grammatical systems, and classified these formalisms on the basis of two features: path complexity; and path independence.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We would like to thank Eugene Charniak, Michael Collins, and Adwait Ratnaparkhi for enabling all of this research by providing us with their parsers and helpful comments.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The core of Yarowsky's algorithm is as follows: where h is defined by the formula in equation 2, with counts restricted to training data examples that have been labeled in step 2.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As we have noted in Section 2, the general semantic class to which a hanzi belongs is often predictable from its semantic radical.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Ends with the feminine affix :: p. 4.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Proper names are assumed to be coreferent if they match exactly, or if they closely match based on a few heuristics.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Table 6: Example Translations for the Verbmobil task.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Gazdar (1985) considers a number of linguistic analyses which IG's (but not CFG's) can make, for example, the Norwedish example shown in Figure 1.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Our clue is the NE instance pairs.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"information structure.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"German English Training: Sentences 58 073 Words 519 523 549 921 Words* 418 979 453 632 Vocabulary Size 7939 4648 Singletons 3454 1699 Test-147: Sentences 147 Words 1 968 2 173 Perplexity { 26:5 Table 4: Multi-reference word error rate (mWER) and subjective sentence error rate (SSER) for three different search procedures.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The search starts in hypothesis (f;g; 0) and ends in the hypotheses (f1; ; Jg; j), with j 2 f1; ; Jg.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"This design leads to a significant reduction in the computational complexity of training and inference.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Here we use a slightly different notion of lift, applying to individual arcs and moving their head upwards one step at a time: Intuitively, lifting an arc makes the word wk dependent on the head wi of its original head wj (which is unique in a well-formed dependency graph), unless wj is a root in which case the operation is undefined (but then wj \u2014* wk is necessarily projective if the dependency graph is well-formed).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"3.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A dynamic programming recursion similar to the one in Eq. 2 is evaluated.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The ATB has a much higher fraction of nuclei per tree, and a higher type-level error rate.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Exploiting Diversity in Natural Language Processing: Combining Parsers","label":1},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The input for the segmentation task is however highly ambiguous for Semitic languages, and surface forms (tokens) may admit multiple possible analyses as in (BarHaim et al., 2007; Adler and Elhadad, 2006).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"One is that smaller sets sometime have meaningless keywords, like \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d in the CC-domain, or \u00e2\u0080\u009ccompare\u00e2\u0080\u009d in the PC-domain.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The performance of our system on those sentences ap\u00c2\u00ad peared rather better than theirs.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Formally, we define dependency graphs as follows: 3.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"They contain about 200M words (25M, 110M, 40M and 19M words, respectively).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The samples from each corpus were independently evaluated.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The annotator can then \u00e2\u0080\u009cclick away\u00e2\u0080\u009d those words that are here not used as connectives (such as the conjunction und (\u00e2\u0080\u0098and\u00e2\u0080\u0099) used in lists, or many adverbials that are ambiguous between connective and discourse particle).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In IE, creating the patterns which express the requested scenario, e.g. \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d or \u00e2\u0080\u009ccorporate merger and acquisition\u00e2\u0080\u009d is regarded as the hardest task.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Same for . In the case where the next token is a hyphen, then is also used as a feature: (init- Caps, ) is set to 1.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Only tokens with initCaps not found in commonWords are tested against each list in Table 2.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The confidence level is then used as the belief value for the knowledge source.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In this section, we try to compare our results with those obtained by IdentiFinder ' 97 (Bikel et al., 1997), IdentiFinder ' 99 (Bikel et al., 1999), and MENE (Borthwick, 1999).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"76 16.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"(If the TF\/IDF score of that word is below a threshold, the phrase is discarded.)","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The algorithm builds two classifiers in parallel from labeled and unlabeled data.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"1 The apparent difficulty of adapting constituency models to non-configurational languages has been one motivation for dependency representations (Hajic\u00cb\u0087 and Zema\u00c2\u00b4nek, 2004; Habash and Roth, 2009).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The importance of dictionaries in NERs has been investigated in the literature (Mikheev et al., 1999).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"08 84.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Table 1 briefly describes the seven syntactic heuristics used by BABAR to resolve noun phrases.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"rhetorical analysis We are experimenting with a hybrid statistical and knowledge-based system for discourse parsing and summarization (Stede 2003), (Hanneforth et al. 2003), again targeting the genre of commentaries.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Time for Moses itself to load, including loading the language model and phrase table, is included.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"It is worth noting that the middle words of the Italian trigrams are nouns too, which exhibits the fact that the similarity metric connects types having the same syntactic category.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We present KenLM, a library that implements two data structures for efficient language model queries, reducing both time and costs.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Le m\u00b4edicament de r\u00b4ef\u00b4erence de Silapo est EPREX\/ERYPO, qui contient de l\u2019\u00b4epo\u00b4etine alfa.","label":0},{"source_text":"There is no global pruning.","target_text":"Here, the pruning threshold t0 = 10:0 is used.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Note that these observa sider suffix features, capitalization features, punctuation, and digit features.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"2 62.2 39.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Consider for a set of constituents the isolated constituent precision parser metric, the portion of isolated constituents that are correctly hypothesized.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Figure 2 shows an excerpt of a sentence from the Italian test set and the tags assigned by four different models, as well as the gold tags.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Previous work deals with this problem by correcting inconsistencies between the named entity classes assigned to different occurrences of the same entity (Borthwick, 1999; Mikheev et al., 1998).","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"In focus is in particular the correlation with rhetorical structure, i.e., the question whether specific rhetorical relations \u00e2\u0080\u0094 or groups of relations in particular configurations \u00e2\u0080\u0094 are signalled by speakers with prosodic means.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"We have presented a new method for non-projective dependency parsing, based on a combination of data-driven projective dependency parsing and graph transformation techniques.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Other strategies could readily 6 As a reviewer has pointed out, it should be made clear that the function for computing the best path is. an instance of the Viterbi algorithm.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"1).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"There is a (costless) transition between the NC node and f,.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This solution also obviates the need to perform word sense disambiguation.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The first setting uses the European Medicines Agency (EMEA) corpus (Tiedemann, 2009) as IN, and the Europarl (EP) corpus (www.statmt.org\/europarl) as OUT, for English\/French translation in both directions.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Motivated by these questions, we significantly raise baselines for three existing parsing models through better grammar engineering.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"It may seem surprising to some readers that the interhuman agreement scores reported here are so low.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"4 69.0 51.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We report results for the best and median hyperparameter settings obtained in this way.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Interpolation search is therefore a form of binary search with better estimates informed by the uniform key distribution.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Once again we present both a non-parametric and a parametric technique for this task.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"63 81.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The word joining is done on the basis of a likelihood criterion.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We consider properties of the tree sets generated by CFG's, Tree Adjoining Grammars (TAG's), Head Grammars (HG's), Categorial Grammars (CG's), and IG's.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"They first collect the NE instance pairs and contexts, just like our method.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(3)), with one term for each classifier.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"TPT Germann et al. (2009) describe tries with better locality properties, but did not release code.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Automatic paraphrase discovery is an important but challenging task.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We use two common techniques, hash tables and sorted arrays, describing each before the model that uses the technique.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This highly effective approach is not directly applicable to the multinomial models used for core SMT components, which have no natural method for combining split features, so we rely on an instance-weighting approach (Jiang and Zhai, 2007) to downweight domain-specific examples in OUT.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Table 1 briefly describes the seven syntactic heuristics used by BABAR to resolve noun phrases.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"While the linear precedence of segmental morphemes within a token is subject to constraints, the dominance relations among their mother and sister constituents is rather free.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"They return a value in the range [0,1], where 0 indicates neutrality and 1 indicates the strongest belief that the candidate and anaphor are coreferent.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Bresnan, Kaplan, Peters, and Zaenen (1982) argue that these structures are needed to describe crossed-serial dependencies in Dutch subordinate clauses.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"As the reviewer also points out, this is a problem that is shared by, e.g., probabilistic context-free parsers, which tend to pick trees with fewer nodes.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Here, $ is the sentence boundary symbol, which is thought to be at position 0 in the target sentence.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"2.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The type-level tag assignments T generate features associated with word types W . The tag assignments constrain the HMM emission parameters \u00ce\u00b8.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The Hebrew token \u2018bcl\u20191, for example, stands for the complete prepositional phrase 'We adopt here the transliteration of (Sima\u2019an et al., 2001).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The phrases have to be the expressions of length less than 5 chunks, appear between two NEs.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Note that the good classes JADE, GOLD and GRASS have lower costs than the bad classes SICKNESS, DEATH and RAT, as desired, so the trend observed for the results of this method is in the right direction.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Some of these approaches (e.g., Lin, Chiang, and Su [1993]) attempt to identify unknown words, but do not ac\u00c2\u00ad tually tag the words as belonging to one or another class of expression.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"As with personal names, we also derive an estimate from text of the probability of finding a transliterated name of any kind (PTN).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"AdaBoost.MH maintains a distribution over instances and labels; in addition, each weak-hypothesis outputs a confidence vector with one confidence value for each possible label.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The model was parameterized such that the joint probability of a (label, feature-set) pair P(yi, xi) is written as The model assumes that (y, x) pairs are generated by an underlying process where the label is first chosen with some prior probability P(yi); the number of features mi is then chosen with some probability P(mi); finally the features are independently generated with probabilities P(xulyi).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Bikel et al.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Memory-based classifiers for the experiments were created using TiMBL (Daelemans et al., 2003).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"La ng ua ge # To ke ns # W or d Ty pe s # Ta gs E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 1 1 7 3 7 6 6 9 4 3 8 6 2 0 3 5 6 8 6 9 9 6 0 5 2 0 6 6 7 8 8 9 3 3 4 1 9 1 4 6 7 4 9 2 0 6 1 8 3 5 6 2 8 3 9 3 7 2 3 2 5 2 8 9 3 1 1 6 4 5 8 2 0 0 5 7 4 5 2 5 1 2 5 4 2 2 4 7 4 1 Table 2: Statistics for various corpora utilized in experiments.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The tee in Figure 9a can be denoted by t1 i223 where tee substitution is used instead of concatenation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For IE, the system must be able to distinguish between semantically similar noun phrases that play different roles in an event.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We resolve this problem by inserting an entry with probability set to an otherwise-invalid value (\u2212oc).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The entries in this table can be compared with those of Table 3 to see how the performance of the combining techniques degrades in the presence of an inferior parser.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"To optimize this function, we used L-BFGS, a quasi-Newton method (Liu and Nocedal, 1989).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"7 www.cis.upenn.edu\/\u00e2\u0088\u00bcpdtb\/ 8 www.eml-research.de\/english\/Research\/NLP\/ Downloads had to buy a new car.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Finally, we incorporate the instance-weighting model into a general linear combination, and learn weights and mixing parameters simultaneously. where c\u03bb(s, t) is a modified count for pair (s, t) in OUT, u(s|t) is a prior distribution, and y is a prior weight.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The likelihood of the observed data under the model is where P(yi, xi) is defined as in (9).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our motivation for using DempsterShafer is that it provides a well-principled framework for combining evidence from multiple sources with respect to competing hypotheses.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"On the other hand, in a translation system one probably wants to treat this string as a single dictionary word since it has a conventional and somewhat unpredictable translation into English.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"2 70.7 52.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Lexicon Feature: The string of the token is used as a feature.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We have shown that, at least given independent human judgments, this is not the case, and that therefore such simplistic measures should be mistrusted.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Step 4.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"This is not an oversight.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"This leads to word- and constituent-boundaries discrepancy, which breaks the assumptions underlying current state-of-the-art statistical parsers.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"3.2 Reordering with IBM Style.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We also removed the context-type feature type when using the CoBoost approach.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For the moment we will assume that there are only two possible labels: each y, is in { \u20141, +1}.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The learned patterns are then normalized and applied to the corpus.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"This suggests that different types of anaphora may warrant different treatment: definite NP resolution may depend more on lexical semantics, while pronoun resolution may depend more on contextual semantics.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"While the proportion of sentences containing non-projective dependencies is often 15\u201325%, the total proportion of non-projective arcs is normally only 1\u20132%.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"For each trigram type x2 x3 x4 in a sequence x1 x2 x3 x4 x5, we count how many times that trigram type co-occurs with the different instantiations of each concept, and compute the point-wise mutual information (PMI) between the two.5 The similarity between two trigram types is given by summing over the PMI values over feature instantiations that they have in common.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"In the maximum entropy framework, there is no such constraint.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Nonetheless, the prospect of a network of annotated discourse resources seems particularly promising if not only a single annotation layer is used but a whole variety of them, so that a systematic search for correlations between them becomes possible, which in turn can lead to more explanatory models of discourse structure.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"This line of work has been motivated by empirical findings that the standard EM-learned unsupervised HMM does not exhibit sufficient word tag sparsity.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"This allows an unbounded amount of information about two separate paths (e.g. an encoding of their length) to be combined and used to influence the later derivation.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The two new terms force the two classifiers to agree, as much as possible, on the unlabeled examples.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Based on these high-confidence alignments we can extract tuples of the form [u H v], where u is a foreign trigram type, whose middle word aligns to an English word type v. Our bilingual similarity function then sets the edge weights in proportion to these tuple counts.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Tsarfaty and Sima\u2019an (2007) have reported state-of-the-art results on Hebrew unlexicalized parsing (74.41%) albeit assuming oracle morphological segmentation.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In the last example, the less restrictive IbmS word reordering leads to a better translation, although the QmS translation is still acceptable.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The accuracy of the sets in representing paraphrase ranged from 73% to 99%, depending on the NE categories and set sizes; the accuracy of the links for two evaluated domains was 73% and 86%.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Mikheev et al.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Also there are cases where one of the two NEs belong to a phrase outside of the relation.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"87 Table 7: Test set results.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"This PP modifies another NP, whose head is a singular noun.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"IRSTLM\u2019s quantized variant is the inspiration for our quantized variant.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Finally, we provide a realistic eval uation in which segmentation is performed both in a pipeline and jointly with parsing (\u00c2\u00a76).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"It differs from previous machine learning-based NERs in that it uses information from the whole document to classify each word, with just one classifier.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"English was again paired with German, French, and Spanish.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The possible analyses of a surface token pose constraints on the analyses of specific segments.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This WFST is then summed with the WFST implementing the dictionary and morphological rules, and the transitive closure of the resulting transducer is computed; see Pereira, Riley, and Sproat (1994) for an explanation of the notion of summing WFSTs.12 Conceptual Improvements over Chang et al.'s Model.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For example, we can easily imagine that the number of paraphrases for \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d is enormous and it is not possible to create a comprehensive inventory by hand.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Search t0 CPU time #search mWER Method [sec] error [%] QmS 0.0 0.07 108 42:6 1.0 0.13 85 37:8 2.5 0.35 44 36:6 5.0 1.92 4 34:6 10.0 10.6 0 34:5 IbmS 0.0 0.14 108 43:4 1.0 0.3 84 39:5 2.5 0.8 45 39:1 5.0 4.99 7 38:3 10.0 28.52 0 38:2 Table 6 shows example translations obtained by the three different approaches.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"This distributional sparsity of syntactic tags is not unique to English 1 The source code for the work presented in this paper is available at http:\/\/groups.csail.mit.edu\/rbg\/code\/typetagging\/.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"02 99.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In Semitic languages the situation is very different.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Recent results (e.g., (Yarowsky 95; Brill 95; Blum and Mitchell 98)) have suggested that unlabeled data can be used quite profitably in reducing the need for supervision.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We use w erations of sampling (see Figure 2 for a depiction).","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Thus we are interested not in extraction, but actual generation from representations that may be developed to different degrees of granularity.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"5 60.6 Table 3: Multilingual Results: We report token-level one-to-one and many-to-one accuracy on a variety of languages under several experimental settings (Section 5).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For previously unseen hanzi in given names, Chang et al. assign a uniform small cost; but we know that some unseen hanzi are merely acci\u00c2\u00ad dentally missing, whereas others are missing for a reason-for example, because they have a bad connotation.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"All three curves remain steep at the maximum training set size of 18818 trees.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This assumption, however, is not inherent to type-based tagging models.","label":1},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Linguistic intuitions like those in the previous section inform language-specific annotation choices.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Once again we present both a non-parametric and a parametric technique for this task.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Again, this deserves further investigation.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Tables 4 and 5 show BABAR\u00e2\u0080\u0099s performance when just one contextual role knowledge source is used at a time.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"MUC7 has also seen hybrids of statistical NERs and hand-coded systems (Mikheev et al., 1998; Borthwick, 1999), notably Mikheev' s system, which achieved the best performance of 93.39% on the official NE test data.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We will show that by giving the first model some global features, MENERGI outperforms Borthwick' s reference resolution classifier.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This intuition is born out by the experimental results.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Trying to integrate constituent ordering and choice of referring expressions, (Chiarcos 2003) developed a numerical model of salience propagation that captures various factors of author\u00e2\u0080\u0099s intentions and of information structure for ordering sentences as well as smaller constituents, and picking appropriate referring expressions.10 Chiarcos used the PCC annotations of co-reference and information structure to compute his numerical models for salience projection across the generated texts.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The performance of our system on those sentences ap\u00c2\u00ad peared rather better than theirs.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Figure 1 shows sample sentences from these domains, which are widely divergent.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Discriminative Instance Weighting for Domain Adaptation in Statistical Machine Translation","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The negative logarithm of t0 is reported.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Input: Ja , wunderbar . Konnen wir machen . MonS: Yes, wonderful.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The transitive closure of the dictionary in (a) is composed with Id(input) (b) to form the WFST (c).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Put another way, the minimum of Equ.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Each of the constituents must have received at least 1 votes from the k parsers, so a > I1 and 2 \u2014 2k\u00b1-1 b > ri-5-111.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Nonetheless, the results of the comparison with human judges demonstrates that there is mileage being gained by incorporating models of these types of words.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In this work, we take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"For each feature type f and tag t, a multinomial \u00cf\u0088tf is drawn from a symmetric Dirichlet distribution with concentration parameter \u00ce\u00b2.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(1992).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"21 84.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Second, comparisons of different methods are not meaningful unless one can eval\u00c2\u00ad uate them on the same corpus.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The English data comes from the WSJ portion of the Penn Treebank and the other languages from the training set of the CoNLL-X multilingual dependency parsing shared task.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Rather we believe several methods have to be developed using different heuristics to discover wider variety of paraphrases.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"A named entity recognizer (NER) is useful in many NLP applications such as information extraction, question answering, etc. On its own, a NER can also provide users who are looking for person or organization names with quick information.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Our full model outperforms the \u201cNo LP\u201d setting because it has better vocabulary coverage and allows the extraction of a larger set of constraint features.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Its correct antecedent is \u00e2\u0080\u009ca revolver\u00e2\u0080\u009d, which is extracted by the caseframe \u00e2\u0080\u009ckilled with <NP>\u00e2\u0080\u009d.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For example, ... fraud related to work on a federally funded sewage plant in Georgia In this case, Georgia is extracted: the NP containing it is a complement to the preposition in; the PP headed by in modifies the NP a federally funded sewage plant, whose head is the singular noun plant.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"A better approach would be to distin guish between these cases, possibly by drawing on the vast linguistic work on Arabic connectives (AlBatal, 1990).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(c) Coordination ambiguity is shown in dependency scores by e.g., \u00e2\u0088\u0097SSS R) and \u00e2\u0088\u0097NP NP NP R).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"For simplicity, we assume that OUT is homogeneous.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In this paper we have argued that Chinese word segmentation can be modeled ef\u00c2\u00ad fectively using weighted finite-state transducers.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Our final average POS tagging accuracy of 83.4% compares very favorably to the average accuracy of Berg-Kirkpatrick et al.\u2019s monolingual unsupervised state-of-the-art model (73.0%), and considerably bridges the gap to fully supervised POS tagging performance (96.6%).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Now assume we have n pairs (xi,, x2,i) drawn from X1 X X2, where the first m pairs have labels whereas for i = m+ 1...n the pairs are unlabeled.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"(Kehler, 1997) also used a DempsterShafer model to merge evidence from different sources for template-level coreference.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"To investigate the influence of these factors, we analyze Modern Standard Arabic (henceforth MSA, or simply \u00e2\u0080\u009cArabic\u00e2\u0080\u009d) because of the unusual opportunity it presents for comparison to English parsing results.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The complexity of the quasimonotone search is O(E3 J (R2+LR)).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"This result suggests the benefit of using the automatic discovery method.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"We have tested the translation system on the Verbmobil task (Wahlster 1993).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Floating point values may be stored in the trie exactly, using 31 bits for non-positive log probability and 32 bits for backoff5.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Because b is a function, no additional hypothesis splitting happens.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We include a list of per-category results for selected phrasal labels, POS tags, and dependencies in Table 8.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The first experiment uses data from two dependency treebanks.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Even with vocalization, there are linguistic categories that are difficult to identify without semantic clues.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The only supervision is in the form of 7 seed rules (namely, that New York, California and U.S. are locations; that any name containing Mr is a person; that any name containing Incorporated is an organization; and that I.B.M. and Microsoft are organizations).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The particular classifier used depends upon the noun.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"On the MUC6 data, Bikel et al.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We weight edges using a unigram language model estimated with Good- Turing smoothing.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"In a fully supervised setting, the task is to learn a function f such that for all i = 1...m, f (xi,i, 12,i) = yz.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Moreover, if we want to test on a huge test corpus, indexing the whole corpus might prove computationally expensive.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The string pumping lemma for CFG's (uvwxy-theorem) can be seen as a corollary of this lemma. from this pumping lemma: a single path can be pumped independently.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This FSA I can be segmented into words by composing Id(I) with D*, to form the WFST shown in Figure 2(c), then selecting the best path through this WFST to produce the WFST in Figure 2(d).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"However, reads in the TRIE data structure are more expensive due to bit-level packing, so we found that it is faster to use interpolation search the entire time.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Some approaches depend upon some form of con\u00c2\u00ad straint satisfaction based on syntactic or semantic features (e.g., Yeh and Lee [1991], which uses a unification-based approach).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"There are simply not enough votes remaining to allow any of the crossing structures to enter the hypothesized constituent set.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"37 79.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"If \u00e2\u0080\u009cgun\u00e2\u0080\u009d and \u00e2\u0080\u009crevolver\u00e2\u0080\u009d refer to the same object, then it should also be acceptable to say that Fred was \u00e2\u0080\u009ckilled with a gun\u00e2\u0080\u009d and that the burglar \u00e2\u0080\u009cfireda revolver\u00e2\u0080\u009d.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"It is unique, agrees with the maximum-likelihood distribution, and has the exponential form (Della Pietra et al., 1997): where refers to the outcome, the history (or context), and is a normalization function.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The first modification \u2014 cautiousness \u2014 is a relatively minor change.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"(1997; 1999) do have some statistics that show how IdentiFinder performs when the training data is reduced.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The constituent voting and na\u00efve Bayes techniques are equivalent because the parameters learned in the training set did not sufficiently discriminate between the three parsers.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"There is a guarantee of no crossing brackets but there is no guarantee that a constituent in the tree has the same children as it had in any of the three original parses.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The highestorder N-gram array omits backoff and the index, since these are not applicable.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Table 4 shows how much the Bayes switching technique uses each of the parsers on the test set.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"yu2 'fish.'","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Precision.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Despite these limitations, a purely finite-state approach to Chinese word segmentation enjoys a number of strong advantages.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"By design, they readily capture regularities at the token-level.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"logical rules, and personal names; the transitive closure of the resulting machine is then computed.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Members of LCFRS whose operations have this property can be translated into the ILFP notation (Rounds, 1985).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"However, in order to capture the properties of various grammatical systems under consideration, our notation is more restrictive that ILFP, which was designed as a general logical notation to characterize the complete class of languages that are recognizable in polynomial time.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We use v1.0 mainly because previous studies on joint inference reported results w.r.t. v1.0 only.5 We expect that using the same setup on v2.0 will allow a crosstreebank comparison.6 We used the first 500 sentences as our dev set and the rest 4500 for training and report our main results on this split.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The paper explains the design decisions taken in the annotations, and describes a number of applications using this corpus with its multi-layer annotation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: Mr. Bush disclosed the policy by reading it...","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"(f1; ;mg n fl1; l2g ; l) 4 (f1; ;m \u00f4\u0080\u0080\u0080 1g n fl1; l2; l3g ; l0) !","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"It was motivated by the observation that the (Yarowsky 95) algorithm added a very large number of rules in the first few iterations.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"When the maSdar lacks a determiner, the constituent as a whole resem bles the ubiquitous annexation construct \u00ef\u00bf\u00bd ?f iDafa.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"3.1 Maximum Entropy.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The terrorism examples reflect fairly obvious relationships: people who are murdered are killed; agents that \u00e2\u0080\u009creport\u00e2\u0080\u009d things also \u00e2\u0080\u009cadd\u00e2\u0080\u009d and \u00e2\u0080\u009cstate\u00e2\u0080\u009d things; crimes that are \u00e2\u0080\u009cperpetrated\u00e2\u0080\u009d are often later \u00e2\u0080\u009ccondemned\u00e2\u0080\u009d.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"As a (crude) approximation, we normalize the extraction patterns with respect to active and passive voice and label those extractions as agents or patients.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"10 Chinese speakers may object to this form, since the suffix f, menD (PL) is usually restricted to.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Considerable amount of work has been done in recent years on the named entity recognition task, partly due to the Message Understanding Conferences (MUC).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In order to handle the necessary word reordering as an optimization problem within our dynamic programming approach, we describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming (Held, Karp, 1962).","label":0},{"source_text":"Here we present two algorithms.","target_text":"Formally, let el (62) be the number of classification errors of the first (second) learner on the training data, and let Eco be the number of unlabeled examples on which the two classifiers disagree.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(2006).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"To evaluate proper-name identification, we randomly se\u00c2\u00ad lected 186 sentences containing 12,000 hanzi from our test corpus and segmented the text automatically, tagging personal names; note that for names, there is always a sin\u00c2\u00ad gle unambiguous answer, unlike the more general question of which segmentation is correct.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"English parsing evaluations usually report results on sentences up to length 40.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"This essentially means that a better grammar tunes the joint model for optimized syntactic disambiguation at least in as much as their hyper parameters do.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Evalb, the standard parsing metric, is biased toward such corpora (Sampson and Babarczy, 2003).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"4.","label":0},{"source_text":"Here we present two algorithms.","target_text":"But we will show that the use of unlabeled data can drastically reduce the need for supervision.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"GL is then used to parse the string tn1 ... tnk_1, where tni is a terminal corresponding to the lattice span between node ni and ni+1.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The distinctions in the ATB are linguistically justified, but complicate parsing.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The ATB annotation distinguishes between verbal and nominal readings of maSdar process nominals.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"BABAR employs information extraction techniques to represent and learn role relationships.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We are unaware of prior results for the Stanford parser.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"85 82.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"To facilitate the comparison of our results to those reported by (Cohen and Smith, 2007) we use their data set in which 177 empty and \u201cmalformed\u201d7 were removed.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This causes a problem for reverse trie implementations, including SRILM itself, because it leaves n+1-grams without an n-gram node pointing to them.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"One side of the decision making process is when we choose to believe a constituent should be in the parse, even though only one parser suggests it.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The Grammar Our parser looks for the most likely tree spanning a single path through the lattice of which the yield is a sequence of lexemes.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Automatic paraphrase discovery is an important but challenging task.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The implementation is In a similar effort, (G\u00c2\u00a8otze 2003) developed a proposal for the theory-neutral annotation of information structure (IS) \u00e2\u0080\u0094 a notoriously difficult area with plenty of conflicting and overlapping terminological conceptions.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Future work along these lines will incorporate other layers of annotation, in particular the syntax information.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We obtained positive results using a very simple phrase-based system in two different adaptation settings: using English\/French Europarl to improve a performance on a small, specialized medical domain; and using non-news portions of the NIST09 training material to improve performance on the news-related corpora.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"With the additional assumption that (s, t) can be restricted to the support of co(s, t), this is equivalent to a \u201cflat\u201d alternative to (6) in which each non-zero co(s, t) is set to one.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"In each figure the upper graph shows the isolated constituent precision and the bottom graph shows the corresponding number of hypothesized constituents.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"An additional case of super-segmental morphology is the case of Pronominal Clitics.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Certainly these linguistic factors increase the difficulty of syntactic disambiguation.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"37.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Ltd., then organization will be more probable.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Using structural information As was explained in the results section, we extracted examples like \u00e2\u0080\u009cSmith estimates Lotus\u00e2\u0080\u009d, from a sentence like \u00e2\u0080\u009cMr.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Step 2.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"newspaper material, but also including kungfu fiction, Buddhist tracts, and scientific material.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The second modification is more important, and is discussed in the next section.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In the numerator, however, the counts of ni1s are quite irregular, in\u00c2\u00ad cluding several zeros (e.g., RAT, none of whose members were seen).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"2 56.2 32.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"In the ATB, :: b asta\u00e2\u0080\u0099adah is tagged 48 times as a noun and 9 times as verbal noun.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"4 7 . 3 8 . 9 2 8 . 8 2 0 . 7 3 2 . 3 3 5 . 2 2 9 . 6 2 7 . 6 1 4 . 2 4 2 . 8 4 5 . 9 4 4 . 3 6 0 . 6 6 1 . 5 4 9 . 9 3 3 . 9 Table 6: Type-level Results: Each cell report the type- level accuracy computed against the most frequent tag of each word type.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Intuitively, as suggested by the example in the introduction, this is the right granularity to capture domain effects.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"A tree set may be said to have dependencies between paths if some &quot;appropriate&quot; subset can be shown to have dependent paths as defined above.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"We believe it is natural for authors to use abbreviations in subsequent mentions of a named entity (i.e., first \u00e2\u0080\u009cPresident George Bush\u00e2\u0080\u009d then \u00e2\u0080\u009cBush\u00e2\u0080\u009d).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"1 1 0.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"A different but supplementary perspective on discourse-based information structure is taken 11ventionalized patterns (e.g., order of informa by one of our partner projects, which is inter tion in news reports).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"1 61.7 37.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For example, in .., says Mr. Cooper, a vice president of.. both a spelling feature (that the string contains Mr.) and a contextual feature (that president modifies the string) are strong indications that Mr. Cooper is of type Person.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Given that part-of-speech labels are properties of words rather than morphemes, it follows that one cannot do part-of-speech assignment without having access to word-boundary information.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Furthermore, we do not connect the English vertices to each other, but only to foreign language vertices.4 The graph vertices are extracted from the different sides of a parallel corpus (De, Df) and an additional unlabeled monolingual foreign corpus Ff, which will be used later for training.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"To evaluate the performance on the segmentation task, we report SEG, the standard harmonic means for segmentation Precision and Recall F1 (as defined in Bar-Haim et al. (2005); Tsarfaty (2006)) as well as the segmentation accuracy SEGTok measure indicating the percentage of input tokens assigned the correct exact segmentation (as reported by Cohen and Smith (2007)).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"IdentiFinder ' 99' s results are considerably better than IdentiFinder ' 97' s. IdentiFinder' s performance in MUC7 is published in (Miller et al., 1998).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Thus, rather than give a single evaluative score, we prefer to compare the performance of our method with the judgments of several human subjects.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(6), with W+ > W_.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"They are set to fixed constants.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.2.2 The Caseframe Network The first type of contextual role knowledge that BABAR learns is the Caseframe Network (CFNet), which identifies caseframes that co-occur in anaphor\/antecedent resolutions.","label":1},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"There is no global pruning.","target_text":"This measure has the advantage of being completely automatic.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We will report the evaluation results in the next subsection.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Both (Tsarfaty, 2006; Cohen and Smith, 2007) have shown that a single integrated framework outperforms a completely streamlined implementation, yet neither has shown a single generative model which handles both tasks.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"While the linear precedence of segmental morphemes within a token is subject to constraints, the dominance relations among their mother and sister constituents is rather free.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"set was based on an earlier version of the Chang et a!.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The tee pumping lemma states that if there is tree, t = 22 t2t3, generated by a CFG G, whose height is more than a predetermined bound k, then all trees of the form ti tP3 for each i > 0 will also generated by G (as shown in Figure 9b).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Chris Dyer integrated the code into cdec.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"However, there will remain a large number of words that are not readily adduced to any produc\u00c2\u00ad tive pattern and that would simply have to be added to the dictionary.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Methods for expanding the dictionary include, of course, morphological rules, rules for segmenting personal names, as well as numeral sequences, expressions for dates, and so forth (Chen and Liu 1992; Wang, Li, and Chang 1992; Chang and Chen 1993; Nie, Jin, and Hannan 1994).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In section 4 we evaluate these transformations with respect to projectivized dependency treebanks, and in section 5 they are applied to parser output.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We model each parse as the decisions made to create it, and model those decisions as independent events.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The substructures that are unanimously hypothesized by the parsers should be preserved after combination, and the combination technique should not foolishly create substructures for which there is no supporting evidence.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Also, in Information Extraction (IE), in which the system tries to extract elements of some events (e.g. date and company names of a corporate merger event), several event instances from different news articles have to be aligned even if these are expressed differently.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Table 2 shows the features used in the current version of the parser.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We have already mentioned the closely related work by Matsoukas et al (2009) on discriminative corpus weighting, and Jiang and Zhai (2007) on (nondiscriminative) instance weighting.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For example, given a sequence F1G1G2, where F1 is a legal single-hanzi family name, and Plural Nouns X g 0 g \"' X X 0 T!i c\"'.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We have not explored this strategy.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"(1), Pr(eI 1) is the language model, which is a trigram language model in this case.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The search starts in hypothesis (f;g; 0) and ends in the hypotheses (f1; ; Jg; j), with j 2 f1; ; Jg.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For instance, the common \"suffixes,\" -nia (e.g.,.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"One class comprises words derived by productive morphologi\u00c2\u00ad cal processes, such as plural noun formation using the suffix ir, menD.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Even this may be nondeterministic, in case the graph contains several non-projective arcs whose lifts interact, but we use the following algorithm to construct a minimal projective transformation D0 = (W, A0) of a (nonprojective) dependency graph D = (W, A): The function SMALLEST-NONP-ARC returns the non-projective arc with the shortest distance from head to dependent (breaking ties from left to right).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"(Carlson, Marcu 2001) responded to this situation with relatively precise (and therefore long!)","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"8 1 8.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We run the baseline Moses system for the French-English track of the 2011 Workshop on Machine Translation,9 translating the 3003-sentence test set.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The problem is a binary classification problem.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A geometrical progression of language families defined by Weir (1987) involves tree sets with increasingly complex path sets.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"For other languages, we use the CoNLL-X multilingual dependency parsing shared task corpora (Buchholz and Marsi, 2006) which include gold POS tags (used for evaluation).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Previous work that involves the gathering of information from the whole document often uses a secondary classifier, which corrects the mistakes of a primary sentence- based classifier.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"While Berg-Kirkpatrick et al.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Terrorism systems must distinguish between people who perpetrate a crime and people who are victims of a crime.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"3 68.9 50.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In total, across all domains, we kept 13,976 phrases with keywords.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"4.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Along with IRSTLM and TPT, our binary format is memory mapped, meaning the file and in-memory representation are the same.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Papers that use this method or minor variants thereof include Liang (1986), Li et al.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"In such cases we use the non-pruned lattice including all (possibly ungrammatical) segmentation, and let the statistics (including OOV) decide.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"However, it can be noted that the results for the least informative encoding, Path, are almost comparable, while the third encoding, Head, gives substantially worse results for both data sets.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"2.","label":0},{"source_text":"There is no global pruning.","target_text":"However there is no global pruning.","label":1},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Hence we decided to select ten commentaries to form a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099, for which the entire range of annotation levels was realized, so that experiments with multi-level querying could commence.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kinds of cue.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We used C = 1.0 as the L2 regularization constant in (Eq.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"While possible to utilize the feature-based log-linear approach described in Berg-Kirkpatrick et al.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For each token , zero, one, or more of the features in each feature group are set to 1.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Different annotations of the same text are mapped into the same data structure, so that search queries can be formulated across annotation levels.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Richer tag sets have been suggested for modeling morphologically complex distinctions (Diab, 2007), but we find that linguistically rich tag sets do not help parsing.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Head Grammars (HG's), introduced by Pollard (1984), is a formalism that manipulates headed strings: i.e., strings, one of whose symbols is distinguished as the head.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Each xii is a member of X, where X is a set of possible features.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We also removed the context-type feature type when using the CoBoost approach.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR employs information extraction techniques to represent and learn role relationships.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In this paper we have argued that Chinese word segmentation can be modeled ef\u00c2\u00ad fectively using weighted finite-state transducers.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The prediction based on these features is a knearest neighbor classification, using the IB1 algorithm and k = 5, the modified value difference metric (MVDM) and class voting with inverse distance weighting, as implemented in the TiMBL software package (Daelemans et al., 2003).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Instead, we condition on the type-level tag assignments T . Specifically, let St = {i|Ti = t} denote the indices of theword types which have been assigned tag t accord ing to the tag assignments T . Then \u00ce\u00b8t is drawn from DIRICHLET(\u00ce\u00b1, St), a symmetric Dirichlet which only places mass on word types indicated by St. This ensures that each word will only be assigned a single tag at inference time (see Section 4).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(If fewer than n rules have Precision greater than pin, we 3Note that taking tlie top n most frequent rules already makes the method robut to low count events, hence we do not use smoothing, allowing low-count high-precision features to be chosen on later iterations. keep only those rules which exceed the precision threshold.) pm,n was fixed at 0.95 in all experiments in this paper.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The traveling salesman problem is an optimization problem which is defined as follows: given are a set of cities S = s1; ; sn and for each pair of cities si; sj the cost dij > 0 for traveling from city si to city sj . We are looking for the shortest tour visiting all cities exactly once while starting and ending in city s1.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Similarly, there is no compelling evidence that either of the syllables of f.ifflll binllang2 'betelnut' represents a morpheme, since neither can occur in any context without the other: more likely fjfflll binllang2 is a disyllabic morpheme.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"To control for the effect of the HSPELL-based pruning, we also experimented with a morphological analyzer that does not perform this pruning.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Yarowsky 95) describes an algorithm for word-sense disambiguation that exploits redundancy in contextual features, and gives impressive performance.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(a) ;IE shi4 'be' => ;IE;IE shi4bu2-shi4 (be-not-be) 'is it?'","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"immediately by a Romanization into the pinyin transliteration scheme; numerals following each pinyin syllable represent tones.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We introduce several new ideas.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We reduce this to O(log log |A|) time by evenly distributing keys over their range then using interpolation search4 (Perl et al., 1978).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Unsupervised Models for Named Entity Classification Collins","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We proposed an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Because all English vertices are going to be labeled, we do not need to disambiguate them by embedding them in trigrams.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"It is. based on the traditional character set rather than the simplified character set used in Singapore and Mainland China.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The baseline alignment model does not permit that a source word is aligned to two or more target words, e.g. for the translation direction from German toEnglish, the German compound noun 'Zahnarztter min' causes problems, because it must be translated by the two target words dentist's appointment.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The entries in this table can be compared with those of Table 3 to see how the performance of the combining techniques degrades in the presence of an inferior parser.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"RandLM 0.2 (Talbot and Osborne, 2007) stores large-scale models in less memory using randomized data structures.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Gabbard and Kulick (2008) show that there is significant attachment ambiguity associated with iDafa, which occurs in 84.3% of the trees in our development set.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"In all cases, the key is collapsed to its 64-bit hash.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In various dialects of Mandarin certain phonetic rules apply at the word.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Handling Unknown tokens When handling unknown tokens in a language such as Hebrew various important aspects have to be borne in mind.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"4.2 A Sample Segmentation Using Only Dictionary Words Figure 4 shows two possible paths from the lattice of possible analyses of the input sentence B X:\u00c2\u00a5 .:.S:P:l 'How do you say octopus in Japanese?' previously shown in Figure 1.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Note that the sets of possible classifiers for a given noun can easily be encoded on that noun by grammatical features, which can be referred to by finite-state grammatical rules.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The links can solve the problem.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Our coreference resolver performed well in two domains, and experiments showed that each contextual role knowledge source contributed valuable information.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"First, it directly encodes linguistic intuitions about POS tag assignments: the model structure reflects the one-tag-per-word property, and a type- level tag prior captures the skew on tag assignments (e.g., there are fewer unique determiners than unique nouns).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"6 Conclusions.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"For each set, the phrases with bracketed frequencies are considered not paraphrases in the set.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Our results suggest that current parsing models would benefit from better annotation consistency and enriched annotation in certain syntactic configurations.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"For effectively annotating connectives\/scopes, we found that existing annotation tools were not well-suited, for two reasons: \u00e2\u0080\u00a2 Some tools are dedicated to modes of annotation (e.g., tiers), which could only quite un-intuitively be used for connectives and scopes.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kind of cue.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The method halves the error rate in comparison to a method using the labeled examples alone.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The weak hypothesis chosen was then restricted to be a predictor in favor of this label.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"A few pointed out that adequacy should be broken up into two criteria: (a) are all source words covered?","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We do not show the numbers for the Bayes models in Table 2 because the parameters involved were established using this set.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We showed in \u00c2\u00a72 that lexical ambiguity explains the underperformance of these categories.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Given a PCFG grammar G and a lattice L with nodes n1 ... nk, we construct the weighted grammar GL as follows: for every arc (lexeme) l E L from node ni to node nj, we add to GL the rule [l --+ tni, tni+1, ... , tnj_1] with a probability of 1 (this indicates the lexeme l spans from node ni to node nj).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"When finished, the whole material is written into an XML-structured annotation file.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For example, a good classifier would identify Mrs. Frank as a person, Steptoe & Johnson as a company, and Honduras as a location.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Our coreference resolver also incorporates an existential noun phrase recognizer and a DempsterShafer probabilistic model to make resolution decisions.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"termined by the category of the word that follows it.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"The entire treebank is used in the experiment, but only primary dependencies are considered.4 In all experiments, punctuation tokens are included in the data but omitted in evaluation scores.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Finally, we would like to note that it is possible to devise similar algorithms based with other objective functions than the one given in Equ.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In this paper, we will propose an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"These three parsers have given the best reported parsing results on the Penn Treebank Wall Street Journal corpus (Marcus et al., 1993).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"For instance, for out-ofdomain English-French, Systran has the best BLEU and manual scores.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"All notions of word, with the exception of the orthographic word, are as relevant in Chinese as they are in English, and just as is the case in other languages, a word in Chinese may correspond to one or more symbols in the orthog 1 For a related approach to the problem of word-segrnention in Japanese, see Nagata (1994), inter alia..","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We used a standard one-pass phrase-based system (Koehn et al., 2003), with the following features: relative-frequency TM probabilities in both directions; a 4-gram LM with Kneser-Ney smoothing; word-displacement distortion model; and word count.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Like verbs, maSdar takes arguments and assigns case to its objects, whereas it also demonstrates nominal characteristics by, e.g., taking determiners and heading iDafa (Fassi Fehri, 1993).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Mohri [1995]) shows promise for improving this situation.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Its correct antecedent is \u00e2\u0080\u009ca revolver\u00e2\u0080\u009d, which is extracted by the caseframe \u00e2\u0080\u009ckilled with <NP>\u00e2\u0080\u009d.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Specifically, we assume each word type W consists of feature-value pairs (f, v).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"When finished, the whole material is written into an XML-structured annotation file.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Previous work that involves the gathering of information from the whole document often uses a secondary classifier, which corrects the mistakes of a primary sentence- based classifier.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"4.1 The Task and the Corpus.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Acronyms (ACRO): Words made up of all capitalized letters in the text zone will be stored as acronyms (e.g., IBM).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"However, we note that the final conditional estimates p(s|t) from a given phrase table maximize the likelihood of joint empirical phrase pair counts over a word-aligned corpus.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For example, a person\u00e2\u0080\u0099s full name will match with just their last name (e.g., \u00e2\u0080\u009cGeorge Bush\u00e2\u0080\u009d and \u00e2\u0080\u009cBush\u00e2\u0080\u009d), and a company name will match with and without a corporate suffix (e.g., \u00e2\u0080\u009cIBM Corp.\u00e2\u0080\u009d and \u00e2\u0080\u009cIBM\u00e2\u0080\u009d).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We thank Felix Hageloh (Hageloh, 2006) for providing us with this version. proposed in (Tsarfaty, 2006).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"It can be shown that the path set of the tree set generated by a TAG G is a context-free language.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"This work was funded by NSF grant IRI-9502312.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Ablation Analysis We evaluate the impact of incorporating various linguistic features into our model in Table 3.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In (Reitter, Stede 2003) we went a different way and suggested URML5, an XML format for underspecifying rhetorical structure: a number of relations can be assigned instead of a single one, competing analyses can be represented with shared forests.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The latter arcs correspond to OOV words in English.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"[Hasegawa et al. 04] reported only on relation discovery, but one could easily acquire para phrases from the results.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Given the limited number of judgements we received, we did not try to evaluate this.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"In addition we report for each model its performance on goldsegmented input (GS) to indicate the upper bound 11Overt definiteness errors may be seen as a wrong feature rather than as wrong constituent and it is by now an accepted standard to report accuracy with and without such errors. for the grammars\u2019 performance on the parsing task.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For example, in Northern dialects (such as Beijing), a full tone (1, 2, 3, or 4) is changed to a neutral tone (0) in the final syllable of many words: Jll donglgual 'winter melon' is often pronounced donglguaO.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"2.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"For developing these mechanisms, the possibility to feed in hand-annotated information is very useful.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Pseudo-Projective Dependency Parsing","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: The brigade, which attacked ...","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"our full model yields 39.3% average error reduction across languages when compared to the basic configuration (1TW).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(a) 1 \u00c2\u00a7 . ;m t 7 leO z h e 4 pil m a 3 lu 4 sh an g4 bi ng 4 t h i s CL (assi fier) horse w ay on sic k A SP (ec t) 'This horse got sick on the way' (b) 1\u00c2\u00a7: . til y zhe4 tiao2 ma3lu4 hen3 shao3 this CL road very few 'Very few cars pass by this road' :$ chel jinglguo4 car pass by 2.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"If (wi, r, wj) E A, we say that wi is the head of wj and wj a dependent of wi.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"This limits the number of NE category pairs to 2,000 and the number of NE pair instances to 0.63 million.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"For comparison to information-retrieval inspired baselines, eg (L\u00a8u et al., 2007), we select sentences from OUT using language model perplexities from IN.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"00 76.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Examples will usually be accompanied by a translation, plus a morpheme-by-morpheme gloss given in parentheses whenever the translation does not adequately serve this purpose.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"We use a solution to this problem similar to the one presented in (Och et al., 1999), where target words are joined during training.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Out of those 15 links, 4 are errors, namely \u00e2\u0080\u009cbuy - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cpurchase - stake\u00e2\u0080\u009d \u00e2\u0080\u009cacquisition - stake\u00e2\u0080\u009d.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Queries detect the invalid probability, using the node only if it leads to a longer match.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The token-level term is similar to the standard HMM sampling equations found in Johnson (2007).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Diacritics can also be used to specify grammatical relations such as case and gender.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Two general approaches are presented and two combination techniques are described for each approach.","label":1},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We present two algorithms.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Even when there is training data available in the domain of interest, there is often additional data from other domains that could in principle be used to improve performance.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Finally, we would like to note that it is possible to devise similar algorithms based with other objective functions than the one given in Equ.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For MUC6, for example, Corporate- Suffix-List is made up of ltd., associates, inc., co, corp, ltd, inc, committee, institute, commission, university, plc, airlines, co., corp. and Person-Prefix- List is made up of succeeding, mr., rep., mrs., secretary, sen., says, minister, dr., chairman, ms. . For a token that is in a consecutive sequence of init then a feature Corporate-Suffix is set to 1.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The first value reports resident size after loading; the second is the gap between post-loading resident memory and peak virtual memory.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Search CPU time mWER SSER Method [sec] [%] [%] MonS 0:9 42:0 30:5 QmS 10:6 34:4 23:8 IbmS 28:6 38:2 26:2 4.2 Performance Measures.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In IE, creating the patterns which express the requested scenario, e.g. \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d or \u00e2\u0080\u009ccorporate merger and acquisition\u00e2\u0080\u009d is regarded as the hardest task.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":".","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"But for eign learners are often surprised by the verbless predications that are frequently used in Arabic.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Kollege.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"More recently, (Riloff and Jones 99) describe a method they term &quot;mutual bootstrapping&quot; for simultaneously constructing a lexicon and contextual extraction patterns.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In (b) is a plausible segmentation for this sentence; in (c) is an implausible segmentation.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We report results for the best and median hyperparameter settings obtained in this way.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"However, it is desirable if we can separate them.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Tsarfaty (2006) used a morphological analyzer (Segal, 2000), a PoS tagger (Bar-Haim et al., 2005), and a general purpose parser (Schmid, 2000) in an integrated framework in which morphological and syntactic components interact to share information, leading to improved performance on the joint task.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Microsoft\u2019s approach uses dependency trees, others use hierarchical phrase models.","label":0},{"source_text":"Here we present two algorithms.","target_text":"To measure the contribution of each modification, a third, intermediate algorithm, Yarowsky-cautious was also tested.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Maximizing (7) is thus much faster than a typical MERT run. where co(s, t) are the counts from OUT, as in (6).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For natural disasters, BABAR generated 20,479 resolutions: 11,652 from lexical seeding and 8,827 from syntactic seeding.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"4 70.4 46.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"To this end, we picked 100 sentences at random containing 4,372 total hanzi from a test corpus.14 (There were 487 marks of punctuation in the test sentences, including the sentence-final periods, meaning that the average inter-punctuation distance was about 9 hanzi.)","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The \u201cNo LP\u201d model does not outperform direct projection for German and Greek, but performs better for six out of eight languages.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"8 57.3 +F EA TS be st me dia n 50.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"13.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The distinctions in the ATB are linguistically justified, but complicate parsing.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"A non-optimal analysis is shown with dotted lines in the bottom frame.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We used a standard one-pass phrase-based system (Koehn et al., 2003), with the following features: relative-frequency TM probabilities in both directions; a 4-gram LM with Kneser-Ney smoothing; word-displacement distortion model; and word count.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The first probability is estimated from a name count in a text database, and the rest of the probabilities are estimated from a large list of personal names.n Note that in Chang et al.'s model the p(rule 9) is estimated as the product of the probability of finding G 1 in the first position of a two-hanzi given name and the probability of finding G2 in the second position of a two-hanzi given name, and we use essentially the same estimate here, with some modifications as described later on.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"So, we set a threshold that at least two examples are required to build a link.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"This is akin to PoS tags sequences induced by different parses in the setup familiar from English and explored in e.g.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We have modified Moses (Koehn et al., 2007) to keep our state with hypotheses; to conserve memory, phrases do not keep state.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Memory usage in PROBING is high, though SRILM is even larger, so where memory is of concern we recommend using TRIE, if it fits in memory.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Our code has been publicly available and intergrated into Moses since October 2010.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In addition to the tapes required to store the indices, M requires one work tape for splitting the substrings.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"One may argue with these efforts on normalization, and ultimately their value should be assessed by assessing their impact on inter-annotator agreement.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Zone: MUC data contains SGML tags, and a document is divided into zones (e.g., headlines and text zones).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Aside from adding a simple rule to correct alif deletion caused by the preposition J, no other language-specific processing is performed.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"F-measure is the harmonic mean of precision and recall, 2PR\/(P + R).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For our demonstration system, we typically use the pruning threshold t0 = 5:0 to speed up the search by a factor 5 while allowing for a small degradation in translation accuracy.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Model components cascade, so the row corresponding to +FEATS also includes the PRIOR component (see Section 3).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We can model this probability straightforwardly enough with a probabilistic version of the grammar just given, which would assign probabilities to the individual rules.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"On the other hand, in a translation system one probably wants to treat this string as a single dictionary word since it has a conventional and somewhat unpredictable translation into English.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Each trie node is individually allocated and full 64-bit pointers are used to find them, wasting memory.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Extract NE instance pairs with contexts First, we extract NE pair instances with their context from the corpus.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Lemma: If the number of votes required by constituent voting is greater than half of the parsers under consideration the resulting structure has no crossing constituents.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This larger corpus was kindly provided to us by United Informatics Inc., R.O.C. a set of initial estimates of the word frequencies.9 In this re-estimation procedure only the entries in the base dictionary were used: in other words, derived words not in the base dictionary and personal and foreign names were not used.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"It falls short of the \u201cProjection\u201d baseline for German, but is statistically indistinguishable in terms of accuracy.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Therefore, we want state to encode the minimum amount of information necessary to properly compute language model scores, so that the decoder will be faster and make fewer search errors.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The contextual role knowledge that BABAR uses for coreference resolution is derived from this caseframe data.","label":0},{"source_text":"There is no global pruning.","target_text":"We use a solution to this problem similar to the one presented in (Och et al., 1999), where target words are joined during training.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Furthermore, even the size of the dictionary per se is less important than the appropriateness of the lexicon to a particular test corpus: as Fung and Wu (1994) have shown, one can obtain substantially better segmentation by tailoring the lexicon to the corpus to be segmented.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"D o m ai n Li n k ac cu ra cy W N c o v e r a g e C C 7 3 . 3 % 2 \/ 1 1 P C 8 8 . 9 % 2 \/ 8 Table 2.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Our motivation for using DempsterShafer is that it provides a well-principled framework for combining evidence from multiple sources with respect to competing hypotheses.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This is not ideal for some applications, however.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"\u00e2\u0080\u009cThe gun\u00e2\u0080\u009d will be extracted by the caseframe \u00e2\u0080\u009cfired <patient>\u00e2\u0080\u009d.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"When aligning the words in parallel texts (for language pairs like SpanishEnglish, French-English, ItalianGerman,...), we typically observe a strong localization effect.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The following features were used: full-string=x The full string (e.g., for Maury Cooper, full- s tring=Maury_Cooper). contains(x) If the spelling contains more than one word, this feature applies for any words that the string contains (e.g., Maury Cooper contributes two such features, contains (Maury) and contains (Cooper) . allcapl This feature appears if the spelling is a single word which is all capitals (e.g., IBM would contribute this feature). allcap2 This feature appears if the spelling is a single word which is all capitals or full periods, and contains at least one period.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Table 3: Dev set frequencies for the two most significant discourse markers in Arabic are skewed toward analysis as a conjunction.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"As lower frequency examples include noise, we set a threshold that an NE category pair should appear at least 5 times to be considered and an NE instance pair should appear at least twice to be considered.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":".","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Recently, combination techniques have been investigated for part of speech tagging with positive results (van Halteren et al., 1998; Brill and Wu, 1998).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"4 Evaluation Results.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"0n0'i'i0'2&quot;bin242bn I n = 711 + n2 } On the other hand, no linguistic use is made of this general form of composition and Steedman (personal communication) and Steedman (1986) argues that a more limited definition of composition is more natural.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"We hypothesize that modeling morphological information will greatly constrain the set of possible tags, thereby further refining the representation of the tag lexicon.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The sequence of states needed to carry out the word reordering example in Fig.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Output of the learning algorithm: a function h:Xxy [0, 1] where h(x, y) is an estimate of the conditional probability p(y1x) of seeing label y given that feature x is present.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"(S; C; j); Not only the coverage set C and the positions j; j0, but also the verbgroup states S; S0 are taken into account.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This differs from other implementations (Stolcke, 2002; Pauls and Klein, 2011) that use hash tables as nodes in a trie, as explained in the next section.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"With respect to exact match, the improvement is even more noticeable, which shows quite clearly that even if non-projective dependencies are rare on the token level, they are nevertheless important for getting the global syntactic structure correct.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We address the question of whether or not a formalism can generate only structural descriptions with independent paths.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The web-based Annis imports data in a variety of XML formats and tagsets and displays it in a tier-orientedway (optionally, trees can be drawn more ele gantly in a separate window).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Here, an NE instance pair is any pair of NEs separated by at most 4 syntactic chunks; for example, \u00e2\u0080\u009cIBM plans to acquire Lotus\u00e2\u0080\u009d.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Before describing the unsupervised case we first describe the supervised version of the algorithm: Input to the learning algorithm: n labeled examples of the form (xi, y\u201e). y, is the label of the ith example (given that there are k possible labels, y, is a member of y = {1 ... 0). xi is a set of mi features {x,1, Xi2 .","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"\u00e2\u0080\u00a2 Bridging links: the annotator is asked to specify the type as part-whole, cause-effect (e.g., She had an accident.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"8 1 2.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Again, famous place names will most likely be found in the dictionary, but less well-known names, such as 1PM\u00c2\u00b1 R; bu4lang3-shi4wei2-ke4 'Brunswick' (as in the New Jersey town name 'New Brunswick') will not generally be found.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Pseudo-labels are formed by taking seed labels on the labeled examples, and the output of the fixed classifier on the unlabeled examples.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"\u00e2\u0080\u009d).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This is a simple and effective alternative to setting weights discriminatively to maximize a metric such as BLEU.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In this paper we present a stochastic finite-state model for segmenting Chinese text into words, both words found in a (static) lexicon as well as words derived via the above-mentioned productive processes.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The second model (+PRIOR) utilizes the independent prior over type-level tag assignments P (T |\u00cf\u0088).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The features we used can be divided into 2 classes: local and global.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Otherwise, the scope of the search problem shrinks recursively: if A[pivot] < k then this becomes the new lower bound: l +\u2014 pivot; if A[pivot] > k then u +\u2014 pivot.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The work of Rounds (1969) shows that the path sets of trees derived by IG's (like those of TAG's) are context-free languages.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Pseudo-Projective Dependency Parsing","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"J:j:l :zhongl :0.0 ;m,Jlong4 :0.0 (mHHaryg9tltHBI) \u00c2\u00a3: _ADV: 5.98 \u00c2\u00a5 :hua2:o.o E :_NC: 4.41 :mln2:o.o mm : guo2 : 0.0 (RopubllcofChlna) .....,.","label":0},{"source_text":"There is no global pruning.","target_text":"Word Re-ordering and DP-based Search in Statistical Machine Translation","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"In the small window on the left, search queries can be entered, here one for an NP that has been annotated on the co-reference layer as bridging.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The productions of HG's are very similar to those of CFG's except that the operation used must be made explicit.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In the pinyin transliterations a dash(-) separates syllables that may be considered part of the same phonological word; spaces are used to separate plausible phonological words; and a plus sign (+) is used, where relevant, to indicate morpheme boundaries of interest.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"It is also worth pointing out a connection with Daum\u00b4e\u2019s (2007) work that splits each feature into domain-specific and general copies.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"How do additional ambiguities caused by devocalization affect statistical learning?","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Local features are features that are based on neighboring tokens, as well as the token itself.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Particular relations are also consistent with particular hypotheses about the segmentation of a given sentence, and the scores for particular relations can be incremented or decremented depending upon whether the segmentations with which they are consistent are \"popular\" or not.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"13.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"For a trigram language model, the partial hypotheses are of the form (e0; e; C; j).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In order to solve this problem, a parse tree is needed to understand that \u00e2\u0080\u009cLotus\u00e2\u0080\u009d is not the object of \u00e2\u0080\u009cestimates\u00e2\u0080\u009d.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"\u2022 Similarly, when the na\u00efve Bayes classifier is configured such that the constituents require estimated probabilities strictly larger than 0.5 to be accepted, there is not enough probability mass remaining on crossing brackets for them to be included in the hypothesis.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"27 80.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Overall, it gives improvements ranging from 1.1% for German to 14.7% for Italian, for an average improvement of 8.3% over the unsupervised feature-HMM model.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"25 16.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"An example is in (i), where the system fails to group t;,f;?\"$?t!: lin2yang2gang3 as a name, because all three hanzi can in principle be separate words (t;,f; lin2 'wood';?\"$ yang2 'ocean'; ?t!; gang3 'harbor').","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In the remainder of the paper, we outline how a class of Linear Context-Free Rewriting Systems (LCFRS's) may be defined and sketch how semilinearity and polynomial recognition of these systems follows.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In (2a), we want to split the two morphemes since the correct analysis is that we have the adverb :1 cai2 'just,' the modal verb neng2 'be able' and the main verb R: Hke4fu2 'overcome'; the competing analysis is, of course, that we have the noun :1 cai2neng2 'talent,' followed by }'lijke4fu2 'overcome.'","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"com \u00c2\u00a7Cambridge, UK Email: nc201@eng.cam.ac.uk \u00c2\u00a9 1996 Association for Computational Linguistics (a) B ) ( , : & ; ? ' H o w d o y o u s a y o c t o p u s i n J a p a n e s e ? ' (b) P l a u s i b l e S e g m e n t a t i o n I B X I I 1 : & I 0 0 r i 4 w e n 2 z h a n g l y u 2 z e n 3 m e 0 s h u o l ' J a p a n e s e ' ' o c t o p u s ' ' h o w ' ' s a y ' (c) Figure 1 I m p l a u s i b l e S e g m e n t a t i o n [\u00c2\u00a7] lxI 1:&I ri4 wen2 zhangl yu2zen3 me0 shuol 'Japan' 'essay' 'fish' 'how' 'say' A Chinese sentence in (a) illustrating the lack of word boundaries.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We also mark all nodes that dominate an SVO configuration (containsSVO).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"On the MUC6 data, Bikel et al.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"0 57.3 51.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The Potsdam Commentary Corpus","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The number of top-ranked pairs to retain is chosen to optimize dev-set BLEU score.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"There is a guarantee of no crossing brackets but there is no guarantee that a constituent in the tree has the same children as it had in any of the three original parses.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Process statistics are already collected by the kernel (and printing them has no meaningful impact on performance).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We picked two domains, the CC-domain and the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain (PC-domain), for the evaluation, as the entire system output was too large to evaluate.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"And indeed, converging on annotation guidelines is even more difficult than it is with co-reference.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"4.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Because the information flow in our graph is asymmetric (from English to the foreign language), we use different types of vertices for each language.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In our situation, the competing hypotheses are the possible antecedents for an anaphor.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The original OUT counts co(s, t) are weighted by a logistic function w\u03bb(s, t): To motivate weighting joint OUT counts as in (6), we begin with the \u201cideal\u201d objective for setting multinomial phrase probabilities 0 = {p(s|t), dst}, which is the likelihood with respect to the true IN distribution pi(s, t).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The importance of dictionaries in NERs has been investigated in the literature (Mikheev et al., 1999).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"For the LM, adaptive weights are set as follows: where \u03b1 is a weight vector containing an element \u03b1i for each domain (just IN and OUT in our case), pi are the corresponding domain-specific models, and \u02dcp(w, h) is an empirical distribution from a targetlanguage training corpus\u2014we used the IN dev set for this.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Lexicalizing several POS tags improves performance.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The use of weighted transducers in particular has the attractive property that the model, as it stands, can be straightforwardly interfaced to other modules of a larger speech or natural language system: presumably one does not want to segment Chinese text for its own sake but instead with a larger purpose in mind.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The first term in the objective function is the graph smoothness regularizer which encourages the distributions of similar vertices (large wij) to be similar.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We can now add a new weak hypothesis 14 based on a feature in X1 with a confidence value al hl and atl are chosen to minimize the function We now define, for 1 <i <n, the following virtual distribution, As before, Ztl is a normalization constant.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","label":1},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"All features were conjoined with the state z.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In our forth model GTnph we add the definiteness status of constituents following Tsarfaty and Sima\u2019an (2007).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The first value reports use immediately after loading while the second reports the increase during scoring. dBerkeleyLM is written in Java which requires memory be specified in advance.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Among these 32 sets, we found the following pairs of sets which have two or more links.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Here, annotation proceeds in two phases: first, the domains and the units of IS are marked as such.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Personal names such as 00, 3R; zhoulenl-lai2 'Zhou Enlai.'","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"2.2 Contextual Role Knowledge.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(6), with W+ > W_.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Formalisms such as the restricted indexed grammars (Gazdar, 1985) and members of the hierarchy of grammatical systems given by Weir (1987) have independent paths, but more complex path sets.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"For a sentence x and a state sequence z, a first order Markov model defines a distribution: (9) where Val(X) corresponds to the entire vocabulary.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Within this framework, we use features intended to capture degree of generality, including the output from an SVM classifier that uses the intersection between IN and OUT as positive examples.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The CoBoost algorithm described above divides the function Zco into two parts: Zco = 40 + 40.","label":0},{"source_text":"There is no global pruning.","target_text":"For each source word f, the list of its possible translations e is sorted according to p(fje) puni(e), where puni(e) is the unigram probability of the English word e. It is su\u00c3\u0086cient to consider only the best 50 words.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The correct resolution in sentence (b) comes from knowledge that people who are kidnapped are often subsequently released.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"When a comparison against previous results requires additional pre-processing, we state it explicitly to allow for the reader to replicate the reported results.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Gather phrases using keywords Next, we select a keyword for each phrase \u00e2\u0080\u0093 the top-ranked word based on the TF\/IDF metric.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Phrase-level granularity distinguishes our work from previous work by Matsoukas et al (2009), who weight sentences according to sub-corpus and genre membership.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"For the LM, adaptive weights are set as follows: where \u03b1 is a weight vector containing an element \u03b1i for each domain (just IN and OUT in our case), pi are the corresponding domain-specific models, and \u02dcp(w, h) is an empirical distribution from a targetlanguage training corpus\u2014we used the IN dev set for this.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Let H be the set of hanzi, p be the set of pinyin syllables with tone marks, and P be the set of grammatical part-of-speech labels.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Hyperparameter settings are sorted according to the median one-to-one metric over runs.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"(Riloff and Shepherd 97) describe a bootstrapping approach for acquiring nouns in particular categories (such as &quot;vehicle&quot; or &quot;weapon&quot; categories).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"The feature-HMM model works better for all languages, generalizing the results achieved for English by Berg-Kirkpatrick et al. (2010).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Rather we believe several methods have to be developed using different heuristics to discover wider variety of paraphrases.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"\u00e2\u0088\u0097NP NP PP R) and \u00e2\u0088\u0097NP NP ADJP R) are both iDafa attachment.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We will report the evaluation results in the next subsection.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For each label (Per s on, organization and Location), take the n contextual rules with the highest value of Count' (x) whose unsmoothed3 strength is above some threshold pmin.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Then, the German infinitive 'besuchen' and the negation particle 'nicht' are translated.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Here we use a slightly different notion of lift, applying to individual arcs and moving their head upwards one step at a time: Intuitively, lifting an arc makes the word wk dependent on the head wi of its original head wj (which is unique in a well-formed dependency graph), unless wj is a root in which case the operation is undefined (but then wj \u2014* wk is necessarily projective if the dependency graph is well-formed).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Reflexive pronouns with only 1 NP in scope..","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We are unaware of prior results for the Stanford parser.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"For displaying and querying the annoated text, we make use of the Annis Linguistic Database developed in our group for a large research effort (\u00e2\u0080\u0098Sonderforschungsbereich\u00e2\u0080\u0099) revolving around 9 2.7 Information structure.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Pairwise comparison is done using the sign test.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Language models are widely applied in natural language processing, and applications such as machine translation make very frequent queries.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Specifically, we assume each word type W consists of feature-value pairs (f, v).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Until now, all evaluations of Arabic parsing\u00e2\u0080\u0094including the experiments in the previous section\u00e2\u0080\u0094have assumed gold segmentation.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Thus we have some confidence that our own performance is at least as good as that of Chang et al.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(a) IDictionary D I D:d\/0.000 B:b\/0.000 B:b\/0.000 ( b ) ( c ) ( d ) I B e s t P a t h ( I d ( I ) o D * ) I cps:nd4.!l(l() Figure 2 An abstract example illustrating the segmentation algorithm.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"98 15.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"To formalize the approach, we introduce four verbgroup states S: Initial (I): A contiguous, initial block of source positions is covered.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"We group the features used into feature groups.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"That is, given a choice between segmenting a sequence abc into abc and ab, c, the former will always be picked so long as its cost does not exceed the summed costs of ab and c: while; it is possible for abc to be so costly as to preclude the larger grouping, this will certainly not usually be the case.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Note that Wang, Li, and Chang's.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"We first make use of our morphological analyzer to find all segmentation possibilities by chopping off all prefix sequence possibilities (including the empty prefix) and construct a lattice off of them.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The most popular approach to dealing with seg\u00c2\u00ad mentation ambiguities is the maximum matching method, possibly augmented with further heuristics.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In future work, we plan to follow-up on this approach and investigate other ways that contextual role knowledge can be used.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"For alif with hamza, normalization can be seen as another level of devocalization.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"As can be seen in Figure 3, the phrases in the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set include completely different relationships, which are not paraphrases.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Our implementation permits jumping to any n-gram of any length with a single lookup; this appears to be unique among language model implementations.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This is akin to PoS tags sequences induced by different parses in the setup familiar from English and explored in e.g.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In this paper, we describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"basically complete, yet some improvements and extensions are still under way.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"18 We are grateful to ChaoHuang Chang for providing us with this set.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"As a partial solution, for pairs of hanzi that co-occur sufficiently often in our namelists, we use the estimated bigram cost, rather than the independence-based cost.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Once again we present both a non-parametric and a parametric technique for this task.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We use label propagation in two stages to generate soft labels on all the vertices in the graph.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"was done by the participants.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Recent work by Finkel and Manning (2009) which re-casts Daum\u00b4e\u2019s approach in a hierarchical MAP framework may be applicable to this problem.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We present several variations for the lexical component P (T , W |\u00cf\u0088), each adding more complex pa rameterizations.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For the translation experiments, Eq. 2 is recursively evaluated.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Up to now, most IE researchers have been creating paraphrase knowledge (or IE patterns) by hand and for specific tasks.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Put another way, the minimum of Equ.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Finally, note that while most feature concepts are lexicalized, others, such as the suffix concept, are not.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In the second and third translation examples, the IbmS word reordering performs worse than the QmS word reordering, since it can not take properly into account the word reordering due to the German verbgroup.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The Recency KS computes the distance between the candidate and the anaphor relative to its scope.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"This drastic tree manipulation is not appropriate for situations in which we want to assign particular structures to sentences.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"9 66.4 47.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The 2nd block contains the IR system, which was tuned by selecting text in multiples of the size of the EMEA training corpus, according to dev set performance.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"It is a sequence of proper nouns within an NP; its last word Cooper is the head of the NP; and the NP has an appositive modifier (a vice president at S.&P.) whose head is a singular noun (president).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank, leaving only sections 22 and 23 completely untouched during the development of any of the parsers.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The scoping heuristics are based on the anaphor type: for reflexive pronouns the scope is the current clause, for relative pronouns it is the prior clause following its VP, for personal pronouns it is the anaphor\u00e2\u0080\u0099s sentence and two preceding sentences, and for definite NPs it is the anaphor\u00e2\u0080\u0099s sentence and eight preceding sentences.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The task can be considered to be one component of the MUC (MUC-6, 1995) named entity task (the other task is that of segmentation, i.e., pulling possible people, places and locations from text before sending them to the classifier).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"However, their inverted variant implements a reverse trie using less CPU and the same amount of memory7.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This heuristics is used to prune all segmentation possibilities involving \u201clexically improper\u201d segments.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"3.2 Reordering with IBM Style.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Let n be some node labeled X in a tree -y (see Figure 3).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"From here on we will refer to the named-entity string itself as the spelling of the entity, and the contextual predicate as the context.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Naseem et al. (2009) and Snyder et al.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"There are still some open issues to be resolved with the format, but it represents a first step.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Word Head Of Complement POS 1 '01 inna \u00e2\u0080\u009cIndeed, truly\u00e2\u0080\u009d VP Noun VBP 2 '01 anna \u00e2\u0080\u009cThat\u00e2\u0080\u009d SBAR Noun IN 3 01 in \u00e2\u0080\u009cIf\u00e2\u0080\u009d SBAR Verb IN 4 01 an \u00e2\u0080\u009cto\u00e2\u0080\u009d SBAR Verb IN Table 1: Diacritized particles and pseudo-verbs that, after orthographic normalization, have the equivalent surface form 0 an.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The average fluency judgement per judge ranged from 2.33 to 3.67, the average adequacy judgement ranged from 2.56 to 4.13.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Since the segmentation corresponds to the sequence of words that has the lowest summed unigram cost, the segmenter under discussion here is a zeroth-order model.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Unigrams also have 64-bit overhead for vocabulary lookup.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Again we notice that the isolated constituent precision is larger than 0.5 only in those partitions that contain very few samples.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Recall.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We implement two data structures: PROBING, designed for speed, and TRIE, optimized for memory.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Inspection of the data shows that at n = 2500, the two classifiers both give labels on 44,281 (49.2%) of the unlabeled examples, and give the same label on 99.25% of these cases.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The system of Berg-Kirkpatrick et al.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"A large number of rules is needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classi- However, we show that the use of data can reduce the requirements for supervision to just 7 simple &quot;seed&quot; rules.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"German city names are replaced by category markers.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For each domain, we created a semantic dictionary by doing two things.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"8 1 2.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Thus at each iteration the algorithm is forced to pick features for the location, person and organization in turn for the classifier being trained.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"One implementation issue deserves some elaboration.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Language models are widely applied in natural language processing, and applications such as machine translation make very frequent queries.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"This is not an oversight.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The input to AdaBoost is a set of training examples ((xi , yi), , (x\u201e.\u201e yrn)).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Finally, quite a few hanzi are homographs, meaning that they may be pronounced in several different ways, and in extreme cases apparently represent different morphemes: The prenominal modifi\u00c2\u00ad cation marker eg deO is presumably a different morpheme from the second morpheme of \u00c2\u00a7eg mu4di4, even though they are written the same way.4 The second point, which will be relevant in the discussion of personal names in Section 4.4, relates to the internal structure of hanzi.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"If a candidate has a belief value \u00e2\u0089\u00a5 .50, then we select that candidate as the antecedent for the anaphor.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"In addition to the named-entity string (Maury Cooper or Georgia), a contextual predictor was also extracted.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"More details on the memory-based prediction can be found in Nivre et al. (2004) and Nivre and Scholz (2004).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"For each language under consideration, Petrov et al. (2011) provide a mapping A from the fine-grained language specific POS tags in the foreign treebank to the universal POS tags.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We used these three parsers to explore parser combination techniques.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Hence we decided to restrict ourselves to only information from the same document.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Moreover, the Stanford parser achieves the most exact Leaf Ancestor matches and tagging accuracy that is only 0.1% below the Bikel model, which uses pre-tagged input.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"For each feature type f and tag t, a multinomial \u00cf\u0088tf is drawn from a symmetric Dirichlet distribution with concentration parameter \u00ce\u00b2.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Thus, most broad-coverage parsers based on dependency grammar have been restricted to projective structures.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Word Re-ordering and DP-based Search in Statistical Machine Translation","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"(2010) reports the best unsupervised results for English.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In this domain the major scenarios involve the things they agreed on, rather than the mere fact that they agreed.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In MUC6, the best result is achieved by SRA (Krupka, 1995).","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","label":1},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In the present study, we limit ourselves to an algorithmic approach, using a deterministic breadthfirst search.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"In most cases, however, these expansions come with a steep increase in model complexity, with respect to training procedure and inference time.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Our goal is to define a class of formal systems, and show that any member of this class will possess certain attractive properties.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For example, in the sentence that starts with \u00e2\u0080\u009cBush put a freeze on . . .","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"An error count of 0:0 is assigned to a perfect translation, and an error count of 1:0 is assigned to a semantically and syntactically wrong translation.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"with the number of exactly matching guess trees.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Now assume we have n pairs (xi,, x2,i) drawn from X1 X X2, where the first m pairs have labels whereas for i = m+ 1...n the pairs are unlabeled.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"We did have a problem classifying some modified noun phrases where the modified phrase does not represent a qualified or restricted form of the head, like \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, as these are both represented by the keyword \u00e2\u0080\u009cchairman\u00e2\u0080\u009d.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Until now, all evaluations of Arabic parsing\u00e2\u0080\u0094including the experiments in the previous section\u00e2\u0080\u0094have assumed gold segmentation.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"9 65.5 46.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"A search restriction especially useful for the translation direction from German to English is presented.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Ideally we\u00e2\u0080\u0099d like to know the thematic role of each extracted noun phrase, but AutoSlog does not generate thematic roles.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"It is well know that language pairs such as EnglishGerman pose more challenges to machine translation systems than language pairs such as FrenchEnglish.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This smooth guarantees that there are no zeroes estimated.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Chang of Tsinghua University, Taiwan, R.O.C., for kindly providing us with the name corpora.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The task can be considered to be one component of the MUC (MUC-6, 1995) named entity task (the other task is that of segmentation, i.e., pulling possible people, places and locations from text before sending them to the classifier).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"\u00e2\u0080\u0094 similar results have been observed across multiple languages.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"On one end of the spectrum are clustering approaches that assign a single POS tag to each word type (Schutze, 1995; Lamar et al., 2010).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"To differentiate between the coordinating and discourse separator functions of conjunctions (Table 3), we mark each CC with the label of its right sister (splitCC).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"to explore how well we can induce POS tags using only the one-tag-per-word constraint.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"7 Big 5 is the most popular Chinese character coding standard in use in Taiwan and Hong Kong.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Memory-based classifiers for the experiments were created using TiMBL (Daelemans et al., 2003).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"While size of the resulting transducers may seem daunting-the segmenter described here, as it is used in the Bell Labs Mandarin TTS system has about 32,000 states and 209,000 arcs-recent work on minimization of weighted machines and transducers (cf.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"This remains as future work.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"For example, in the phrase \u00e2\u0080\u009cCompany-A last week purchased rival Marshalls from Company-B\u00e2\u0080\u009d, the purchased company is Marshalls, not Company-B.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"If is one of Monday, Tuesday, . . .","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Both BBN and NYU have tagged their own data to supplement the official training data.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Thus corresponding pseudo-labels for instances on which gj abstain are set to zero and these instances do not contribute to the objective function.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Given around 90,000 unlabeled examples, the methods described in this paper classify names with over 91% accuracy.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"First, a non-anaphoric NP classifier identifies definite noun phrases that are existential, using both syntactic rules and our learned existential NP recognizer (Bean and Riloff, 1999), and removes them from the resolution process.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"For example, in TAG's a derived auxiliary tree spans two substrings (to the left and right of the foot node), and the adjunction operation inserts another substring (spanned by the subtree under the node where adjunction takes place) between them (see Figure 3).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Therefore, performance is more closely tied to the underlying data structure than to the cache.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"(Kehler, 1997) also used a DempsterShafer model to merge evidence from different sources for template-level coreference.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Find keywords for each NE pair The keywords are found for each NE category pair.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"In this section, we extend state to optimize left-to-right queries.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"We found that contextual role knowledge was more beneficial for pronouns than for definite noun phrases.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In Section 5, we try to compare results of MENE, IdentiFinder, and MENERGI.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In general, different modalities (\u00e2\u0080\u009cplanned to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cagreed to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cbought\u00e2\u0080\u009d) were considered to express the same relationship within an extraction setting.","label":0},{"source_text":"Here we present two algorithms.","target_text":"It may be more realistic to replace the second criteria with a softer one, for example (Blum and Mitchell 98) suggest the alternative Alternatively, if Ii and 12 are probabilistic learners, it might make sense to encode the second constraint as one of minimizing some measure of the distance between the distributions given by the two learners.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We consider the unsupervised POS induction problem without the use of a tagging dictionary.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The task can be considered to be one component of the MUC (MUC-6, 1995) named entity task (the other task is that of segmentation, i.e., pulling possible people, places and locations from text before sending them to the classifier).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The difference is due to more careful S-NOM NP NP NP VP VBG :: b NP restoring NP ADJP NN :: b NP NN NP NP ADJP DTJJ ADJP DTJJ NN :: b NP NP NP ADJP ADJP DTJJ J ..i NN :: b NP NP NP ADJP ADJP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ DTJJ J ..i _;\u00ef\u00bf\u00bd PRP J ..i NN _;\u00ef\u00bf\u00bd NP PRP DTJJ NN _;\u00ef\u00bf\u00bd NP PRP DTJJ J ..i role its constructive effective (b) Stanford (c) Berkeley (d) Bik el (a) Reference Figure 4: The constituent Restoring of its constructive and effective role parsed by the three different models (gold segmentation).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Adam Pauls provided a pre-release comparison to BerkeleyLM and an initial Java interface.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Co-occurrences among the particles themselves are subject to further syntactic and lexical constraints relative to the stem.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The function A : F \u2014* C maps from the language specific fine-grained tagset F to the coarser universal tagset C and is described in detail in \u00a76.2: Note that when tx(y) = 1 the feature value is 0 and has no effect on the model, while its value is \u2212oc when tx(y) = 0 and constrains the HMM\u2019s state space.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"There, a lattice is used to represent the possible sentences resulting from an interpretation of an acoustic model.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"TPT Germann et al. (2009) describe tries with better locality properties, but did not release code.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Thus we opted not to take the step of creating more precise written annotation guidelines (as (Carlson, Marcu 2001) did for English), which would then allow for measuring inter-annotator agreement.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The sources of our dictionaries are listed in Table 2.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Intuitively, it places more weight on OUT when less evidence from IN is available.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We focus on phrases which connect two Named Entities (NEs), and proceed in two stages.","label":1},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(a) 1 \u00c2\u00a7 . ;m t 7 leO z h e 4 pil m a 3 lu 4 sh an g4 bi ng 4 t h i s CL (assi fier) horse w ay on sic k A SP (ec t) 'This horse got sick on the way' (b) 1\u00c2\u00a7: . til y zhe4 tiao2 ma3lu4 hen3 shao3 this CL road very few 'Very few cars pass by this road' :$ chel jinglguo4 car pass by 2.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The NP is a complement to a preposition, which is the head of a PP.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Nonstochastic lexical-knowledge-based approaches have been much more numer\u00c2\u00ad ous.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"However, 1 http:\/\/maxent.sourceforge.net 3.2 Testing.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"41.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Part of the gap between resident and virtual memory is due to the time at which data was collected.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"In-domain Systran scores on this metric are lower than all statistical systems, even the ones that have much worse human scores.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Note that these observa sider suffix features, capitalization features, punctuation, and digit features.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Global context from the whole document is available and can be exploited in a natural manner with a maximum entropy classifier.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"As can be seen, GR and this \"pared-down\" statistical method perform quite similarly, though the statistical method is still slightly better.16 AG clearly performs much less like humans than these methods, whereas the full statistical algorithm, including morphological derivatives and names, performs most closely to humans among the automatic methods.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In Chinese text, individual characters of the script, to which we shall refer by their traditional name of hanzi,Z are written one after another with no intervening spaces; a Chinese sentence is shown in Figure 1.3 Partly as a result of this, the notion \"word\" has never played a role in Chinese philological tradition, and the idea that Chinese lacks any\u00c2\u00ad thing analogous to words in European languages has been prevalent among Western sinologists; see DeFrancis (1984).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The correct resolution in sentence (c) depends on knowledge that kidnappers frequently blindfold their victims.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"It is generally considered to be fast (Pauls 29 \u2212 1 probabilities and 2' \u2212 2 non-zero backoffs. and Klein, 2011), with a default implementation based on hash tables within each trie node.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"6) are noisy, the results confirm that label propagation within the foreign language part of the graph adds significant quality for every language.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Taking the intersection of languages in these resources, and selecting languages with large amounts of parallel data, yields the following set of eight Indo-European languages: Danish, Dutch, German, Greek, Italian, Portuguese, Spanish and Swedish.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The sequence of states needed to carry out the word reordering example in Fig.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"If there are too many distinct states, the decoder prunes low-scoring partial hypotheses, possibly leading to a search error.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Better grammars are shown here to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"All four of the techniques studied result in parsing systems that perform better than any previously reported.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"For \u00c3\u0086 = 1, a new target language word is generated using the trigram language model p(eje0; e00).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"However, they list two sets, one consisting of 28 fragments and the other of 22 fragments, in which they had 0% recall and precision.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Another approach to finding paraphrases is to find phrases which take similar subjects and objects in large corpora by using mutual information of word distribution [Lin and Pantel 01].","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"It reads a file with a list of German connectives, and when a text is opened for annotation, it highlights all the words that show up in this list; these will be all the potential connectives.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"We can do that . Input: Das ist zu knapp , weil ich ab dem dritten in Kaiserslautern bin . Genaugenommen nur am dritten . Wie ware es denn am ahm Samstag , dem zehnten Februar ? MonS: That is too tight , because I from the third in Kaiserslautern . In fact only on the third . How about ahm Saturday , the tenth of February ? QmS: That is too tight , because I am from the third in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? IbmS: That is too tight , from the third because I will be in Kaiserslautern . In fact only on the third . Ahm how about Saturday , February the tenth ? Input: Wenn Sie dann noch den siebzehnten konnten , ware das toll , ja . MonS: If you then also the seventeenth could , would be the great , yes . QmS: If you could then also the seventeenth , that would be great , yes . IbmS: Then if you could even take seventeenth , that would be great , yes . Input: Ja , das kommt mir sehr gelegen . Machen wir es dann am besten so . MonS: Yes , that suits me perfectly . Do we should best like that . QmS: Yes , that suits me fine . We do it like that then best . IbmS: Yes , that suits me fine . We should best do it like that .","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Table 5: Individual Performance of KSs for Disasters (e.g., \u00e2\u0080\u009cthe mayor\u00e2\u0080\u009d vs. \u00e2\u0080\u009cthe journalist\u00e2\u0080\u009d).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The approach uses both spelling and contextual rules.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"All the sentences have been analyzed by our chunker and NE tag- ger.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Moses sets the cache size parameter to 50 so we did as well; the resulting cache size is 2.82 GB.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"10) and trained both EM and L-BFGS for 1000 iterations.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Firstly, Hebrew unknown tokens are doubly unknown: each unknown token may correspond to several segmentation possibilities, and each segment in such sequences may be able to admit multiple PoS tags.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"(1), Pr(eI 1) is the language model, which is a trigram language model in this case.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In Section 3, we introduce our novel concept to word reordering and a DP-based search, which is especially suitable for the translation direction from German to English.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Thus, the effects of spontaneous speech are present in the corpus, e.g. the syntactic structure of the sentence is rather less restricted, however the effect of speech recognition errors is not covered.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The accuracy of the sets in representing paraphrase ranged from 73% to 99%, depending on the NE categories and set sizes; the accuracy of the links for two evaluated domains was 73% and 86%.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Unigram records store probability, backoff, and an index in the bigram table.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Finally, we incorporate the instance-weighting model into a general linear combination, and learn weights and mixing parameters simultaneously. where c\u03bb(s, t) is a modified count for pair (s, t) in OUT, u(s|t) is a prior distribution, and y is a prior weight.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Table 3: Dev set frequencies for the two most significant discourse markers in Arabic are skewed toward analysis as a conjunction.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"3 These are not full case frames in the traditional sense, but they approximate a simple case frame with a single slot.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The second row is the accuracy of the best of the three parsers.'","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"It is a relatively frequent word in the domain, but it can be used in different extraction scenarios.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We will refer to the semantic classes that co-occur with a caseframe as the semantic expectations of the caseframe.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The choice between different actions is in general nondeterministic, and the parser relies on a memorybased classifier, trained on treebank data, to predict the next action based on features of the current parser configuration.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"immediately by a Romanization into the pinyin transliteration scheme; numerals following each pinyin syllable represent tones.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Here NO counts the number of hypothesized constituents in the development set that match the binary predicate specified as an argument.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The Penn Arabic Treebank (ATB) syntactic guidelines (Maamouri et al., 2004) were purposefully borrowed without major modification from English (Marcus et al., 1993).","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The portions of information in the large window can be individually clicked visible or invisible; here we have chosen to see (from top to bottom) \u00e2\u0080\u00a2 the full text, \u00e2\u0080\u00a2 the annotation values for the activated annotation set (co-reference), \u00e2\u0080\u00a2 the actual annotation tiers, and \u00e2\u0080\u00a2 the portion of text currently \u00e2\u0080\u0098in focus\u00e2\u0080\u0099 (which also appears underlined in the full text).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Ignoring the identity of the target language words e and e0, the possible partial hypothesis extensions due to the IBM restrictions are shown in Table 2.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"0 70.9 42.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The second experiment is limited to data from PDT.5 The training part of the treebank was projectivized under different encoding schemes and used to train memory-based dependency parsers, which were run on the test part of the treebank, consisting of 7,507 sentences and 125,713 tokens.6 The inverse transformation was applied to the output of the parsers and the result compared to the gold standard test set.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The major problem for all segmentation systems remains the coverage afforded by the dictionary and the lexical rules used to augment the dictionary to deal with unseen words.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Furthermore, we know one of the original parses will be the hypothesized parse, so the direct method of determining which one is best is to compute the probability of each of the candidate parses using the probabilistic model we developed in Section 2.1.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Frequency is calculated by counting the number of distinct previous tokens that each token has (e.g., if Electric Corp. is seen 3 times, and Manufacturing Corp. is seen 5 times during training, and Corp. is not seen with any other preceding tokens, then the \u00e2\u0080\u009cfrequency\u00e2\u0080\u009d of Corp. is 2).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This flexibility, along with the simplicity of implementation and expansion, makes this framework an attractive base for continued research.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Email: gale@research.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"As a partial solution, for pairs of hanzi that co-occur sufficiently often in our namelists, we use the estimated bigram cost, rather than the independence-based cost.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In this paper, we show how non-projective dependency parsing can be achieved by combining a datadriven projective parser with special graph transformation techniques.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, management succession systems must distinguish between a person who is fired and a person who is hired.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We have developed a general approach for combining parsers when preserving the entire structure of a parse tree is important.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Cluster phrases based on Links We now have a set of phrases which share a keyword.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Now we have sets of phrases which share a keyword and we have links between those sets.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"For parsing, this is a mistake, especially in the case of interrogatives.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"In both cases, the instanceweighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline, and gains of between 0.6 and 1.8 over an equivalent mixture model (with an identical training procedure but without instance weighting).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Consequently, we implemented our own annotation tool ConAno in Java (Stede, Heintze 2004), which provides specifically the functionality needed for our purpose.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The two knowledge sources that use semantic expectations, WordSemCFSem and CFSemCFSem, always return values of -1 or 0.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If two nouns have mutually exclusive semantic classes, then they cannot be coreferent.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Figure 2: An ATB sample from the human evaluation.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Blum and Mitchell 98) go on to give PAC results for learning in the cotraining case.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":", December, then the feature MonthName is set to 1.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Among these 32 sets, we found the following pairs of sets which have two or more links.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"For t = 1, T and for j = 1, 2: where 4 = exp(-jg'(xj,i)). practice, this greedy approach almost always results in an overall decrease in the value of Zco.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Segmentation errors cascade into the parsing phase, placing an artificial limit on parsing performance.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"For \u00c3\u0086 = 0, no new target word is generated, while an additional source sentence position is covered.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"There is a fairly large body of work on SMT adaptation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"3.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"A secondary reference resolution classifier has information on the class assigned by the primary classifier.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Thus in a two-hanzi word like lflli?J zhong1guo2 (middle country) 'China' there are two syllables, and at the same time two morphemes.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The algorithm builds two classifiers iteratively: each iteration involves minimization of a continuously differential function which bounds the number of examples on which the two classifiers disagree.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Each xt E 2x is the set of features constituting the ith example.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"If one of these checks fails then this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"3) A tight coupling with the speech recognizer output.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"However, as we have noted, nothing inherent in the approach precludes incorporating higher-order constraints, provided they can be effectively modeled within a finite-state framework.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Secondly, some segments in a proposed segment sequence may in fact be seen lexical events, i.e., for some p tag Prf(p \u2014* (s, p)) > 0, while other segments have never been observed as a lexical event before.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"This design leads to a significant reduction in the computational complexity of training and inference.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"It has no syntactic function.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Instead, we focused on phrases and set the frequency threshold to 2, and so were able to utilize a lot of phrases while minimizing noise.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For example, in the sentence that starts with \u00e2\u0080\u009cBush put a freeze on . . .","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"It is also true of the adaptation of the Collins parser for Czech (Collins et al., 1999) and the finite-state dependency parser for Turkish by Oflazer (2003).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"A compatible view is presented by Charniak et al. (1996) who consider the kind of probabilities a generative parser should get from a PoS tagger, and concludes that these should be P(w|t) \u201cand nothing fancier\u201d.3 In our setting, therefore, the Lattice is not used to induce a probability distribution on a linear context, but rather, it is used as a common-denominator of state-indexation of all segmentations possibilities of a surface form.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Any NLP application that presumes as input unrestricted text requires an initial phase of text analysis; such applications involve problems as diverse as machine translation, information retrieval, and text-to-speech synthesis (TIS).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"This withdrawal by the treasury secretary is understandable, though.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Particles are uninflected.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"A summary of the corpus used in the experiments is given in Table 3.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"It also does not prune, so comparing to our pruned model would be unfair.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We were intentionally lenient with our baselines: bilingual information by projecting POS tags directly across alignments in the parallel data.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In this situation, BABAR takes the conservative approach and declines to make a resolution.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"0 57.2 43.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Since trees in a tree set are adjoined together, the addressing scheme uses a sequence of pairings of the address and name of the elementary tree adjoined at that address.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Clearly it is possible to write a rule that states that if an analysis Modal+ Verb is available, then that is to be preferred over Noun+ Verb: such a rule could be stated in terms of (finite-state) local grammars in the sense of Mohri (1993).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The Hebrew token \u2018bcl\u20191, for example, stands for the complete prepositional phrase 'We adopt here the transliteration of (Sima\u2019an et al., 2001).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"At first glance, the problem seems quite complex: a large number of rules is needed to cover the domain, suggesting that a large number of labeled examples is required to train an accurate classifier.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The probabilistic version of this procedure is straightforward: We once again assume independence among our various member parsers.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"In all cases, the key is collapsed to its 64-bit hash.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"For seven out of eight languages a threshold of 0.2 gave the best results for our final model, which indicates that for languages without any validation set, r = 0.2 can be used.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"\u00e2\u0080\u009cAgree\u00e2\u0080\u009d is a subject control verb, which dominates another verb whose subject is the same as that of \u00e2\u0080\u009cagree\u00e2\u0080\u009d; the latter verb is generally the one of interest for extraction.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Our work is motivated by the observation that contextual roles can be critically important in determining the referent of a noun phrase.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"For seven out of eight languages a threshold of 0.2 gave the best results for our final model, which indicates that for languages without any validation set, r = 0.2 can be used.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The POS distributions over the foreign trigram types are used as features to learn a better unsupervised POS tagger (\u00a75).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Process nominals name the action of the transitive or ditransitive verb from which they derive.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For corporate suffixes, a list of tokens cslist that occur frequently as the last token of an organization name is collected from the training data.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(Other classes handled by the current system are discussed in Section 5.)","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Given n-gram counts {cn}Nn=1, we use Flog2 c1] bits per vocabulary identifier and Flog2 cn] per index into the table of ngrams.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Again, famous place names will most likely be found in the dictionary, but less well-known names, such as 1PM\u00c2\u00b1 R; bu4lang3-shi4wei2-ke4 'Brunswick' (as in the New Jersey town name 'New Brunswick') will not generally be found.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Others depend upon various lexical heuris\u00c2\u00ad tics: for example Chen and Liu (1992) attempt to balance the length of words in a three-word window, favoring segmentations that give approximately equal length for each word.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The corpora for both settings are summarized in table 1.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"It has no syntactic function.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Word Re-ordering and DP-based Search in Statistical Machine Translation","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"We believe it is natural for authors to use abbreviations in subsequent mentions of a named entity (i.e., first \u00e2\u0080\u009cPresident George Bush\u00e2\u0080\u009d then \u00e2\u0080\u009cBush\u00e2\u0080\u009d).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Multiple features can be used for the same token.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Next, we represent the input sentence as an unweighted finite-state acceptor (FSA) I over H. Let us assume the existence of a function Id, which takes as input an FSA A, and produces as output a transducer that maps all and only the strings of symbols accepted by A to themselves (Kaplan and Kay 1994).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"\u00e2\u0080\u009cAgree\u00e2\u0080\u009d is a subject control verb, which dominates another verb whose subject is the same as that of \u00e2\u0080\u009cagree\u00e2\u0080\u009d; the latter verb is generally the one of interest for extraction.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We are currently exploring other methods that employ similar ideas and their formal properties.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The first method uses a similar algorithm to that of (Yarowsky 95), with modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"There has also been some work on adapting the word alignment model prior to phrase extraction (Civera and Juan, 2007; Wu et al., 2005), and on dynamically choosing a dev set (Xu et al., 2007).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"On the other hand, given that all schemes have similar parsing accuracy overall, this means that the Path scheme is the least likely to introduce errors on projective arcs.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"As can be seen from the last column in Table 1, both Head and Head+Path may theoretically lead to a quadratic increase in the number of distinct arc labels (Head+Path being worse than Head only by a constant factor), while the increase is only linear in the case of Path.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"As noted in Section 1, our code finds the longest matching entry wnf for query p(wn|s(wn\u22121 f ) The probability p(wn|wn\u22121 f ) is stored with wnf and the backoffs are immediately accessible in the provided state s(wn\u22121 When our code walks the data structure to find wnf , it visits wnn, wnn\u22121, ... , wnf .","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"However, when we pre- tag the input\u00e2\u0080\u0094as is recommended for English\u00e2\u0080\u0094 we notice a 0.57% F1 improvement.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We evaluate our approach on seven languages: English, Danish, Dutch, German, Portuguese, Spanish, and Swedish.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Finally, we would like to note that it is possible to devise similar algorithms based with other objective functions than the one given in Equ.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Step 4.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The weight on each sentence is a value in [0, 1] computed by a perceptron with Boolean features that indicate collection and genre membership.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This leads to a linear combination of domain-specific probabilities, with weights in [0, 1], normalized to sum to 1.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We use the universal POS tagset of Petrov et al. (2011) in our experiments.10 This set C consists of the following 12 coarse-grained tags: NOUN (nouns), VERB (verbs), ADJ (adjectives), ADV (adverbs), PRON (pronouns), DET (determiners), ADP (prepositions or postpositions), NUM (numerals), CONJ (conjunctions), PRT (particles), PUNC (punctuation marks) and X (a catch-all for other categories such as abbreviations or foreign words).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The probabilities are incorporated into the DempsterShafer model using Equation 1.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Our monolingual similarity function (for connecting pairs of foreign trigram types) is the same as the one used by Subramanya et al. (2010).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"An additional case of super-segmental morphology is the case of Pronominal Clitics.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"For instance, on Spanish, the absolute gap on median performance is 10%.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"By design, they readily capture regularities at the token-level.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The translation search is carried out with the category markers and the city names are resubstituted into the target sentence as a postprocessing step.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Our new approach, which is called quasi-monotone search, processes the source sentence monotonically, while explicitly taking into account the positions of the German verbgroup.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Recently, lattices have been used successfully in the parsing of Hebrew (Tsarfaty, 2006; Cohen and Smith, 2007), a Semitic language with similar properties to Arabic.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Also, we don\u00e2\u0080\u0099t know how many such paraphrase sets are necessary to cover even some everyday things or events.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"This suggests a strategy: run interpolation search until the range narrows to 4096 or fewer entries, then switch to binary search.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Again, the idea is that having a picture of syntax, co-reference, and sentence-internal information structure at one\u00e2\u0080\u0099s disposal should aid in finding models of discourse structure that are more explanatory and can be empirically supported.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The first modification \u2014 cautiousness \u2014 is a relatively minor change.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Despite their simplicity, uni- gram weights have been shown as an effective feature in segmentation models (Dyer, 2009).13 The joint parser\/segmenter is compared to a pipeline that uses MADA (v3.0), a state-of-the-art Arabic segmenter, configured to replicate ATB segmentation (Habash and Rambow, 2005).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Figure 1 shows an example.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In the denomi 11 We have two such lists, one containing about 17,000 full names, and another containing frequencies of.","label":0},{"source_text":"Here we present two algorithms.","target_text":"This modification brings the method closer to the DL-CoTrain algorithm described earlier, and is motivated by the intuition that all three labels should be kept healthily populated in the unlabeled examples, preventing one label from dominating \u2014 this deserves more theoretical investigation.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"TPT Germann et al. (2009) describe tries with better locality properties, but did not release code.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Annotation of syntactic structure for the core corpus has just begun.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Finally, we wish to reiterate an important point.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"3 Throughout this paper we shall give Chinese examples in traditional orthography, followed.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Each treelet (an internal node with all its children) represents the use of a rule that is encapsulated by the grammar The grammar encapsulates (either explicitly or implicitly) a finite number of rules that can be written as follows: n > 0 In the case of CFG's, for each production In the case of TAG's, a derivation step in which the derived trees RI, \u2022 \u2022 \u2022 , On are adjoined into fi at rhe addresses \u2022 \u2022 \u2022 \u2022 in. would involve the use of the following rule2.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"\u00ce\u00b8 has a belief value of 1.0, indicating complete certainty that the correct hypothesis is included in the set, and a plausibility value of 1.0, indicating that there is no evidence for competing hypotheses.5 As evidence is collected and the likely hypotheses are whittled down, belief is redistributed to subsets of \u00ce\u00b8.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Figure 4 shows some such phrase sets based on keywords in the CC-domain.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"In line with perplexity results from Table 1, the PROBING model is the fastest followed by TRIE, and subsequently other packages.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Trying to integrate constituent ordering and choice of referring expressions, (Chiarcos 2003) developed a numerical model of salience propagation that captures various factors of author\u00e2\u0080\u0099s intentions and of information structure for ordering sentences as well as smaller constituents, and picking appropriate referring expressions.10 Chiarcos used the PCC annotations of co-reference and information structure to compute his numerical models for salience projection across the generated texts.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Our TRIE implementation is designed to improve upon IRSTLM using a reverse trie with improved search, bit level packing, and stateful queries.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The PROBING data structure is a rather straightforward application of these hash tables to store Ngram language models.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The most frequent NE category pairs are \u00e2\u0080\u009cPerson - Person (209,236), followed by \u00e2\u0080\u009cCountry - Coun- try\u00e2\u0080\u009d (95,123) and \u00e2\u0080\u009cPerson - Country\u00e2\u0080\u009d (75,509).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"More complex approaches such as the relaxation technique have been applied to this problem Fan and Tsai (1988}.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"More details on the memory-based prediction can be found in Nivre et al. (2004) and Nivre and Scholz (2004).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"was done by the participants.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Our initial experimentation with the evaluation tool showed that this is often too overwhelming.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Exploiting Diversity in Natural Language Processing: Combining Parsers","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"We can model this probability straightforwardly enough with a probabilistic version of the grammar just given, which would assign probabilities to the individual rules.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, a person\u00e2\u0080\u0099s full name will match with just their last name (e.g., \u00e2\u0080\u009cGeorge Bush\u00e2\u0080\u009d and \u00e2\u0080\u009cBush\u00e2\u0080\u009d), and a company name will match with and without a corporate suffix (e.g., \u00e2\u0080\u009cIBM Corp.\u00e2\u0080\u009d and \u00e2\u0080\u009cIBM\u00e2\u0080\u009d).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Their best model yields 44.5% one-to-one accuracy, compared to our best median 56.5% result.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Raj and Whittaker (2003) show that integers in a trie implementation can be compressed substantially.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In various dialects of Mandarin certain phonetic rules apply at the word.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The precision and recall measures (described in more detail in Section 3) used in evaluating Treebank parsing treat each constituent as a separate entity, a minimal unit of correctness.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"A simple lexicalized PCFG with second order Markovization gives relatively poor performance: 75.95% F1 on the test set.8 But this figure is surprisingly competitive with a recent state-of-the-art baseline (Table 7).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Equ.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The first step in the learning process is to generate training examples consisting of anaphor\/antecedent resolutions.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"There is a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099 of ten commentaries, for which the range of information (except for syntax) has been completed; the remaining data has been annotated to different degrees, as explained below.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"To measure the contribution of each modification, a third, intermediate algorithm, Yarowsky-cautious was also tested.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Clearly this is not the only way to estimate word-frequencies, however, and one could consider applying other methods: in partic\u00c2\u00ad ular since the problem is similar to the problem of assigning part-of-speech tags to an untagged corpus given a lexicon and some initial estimate of the a priori probabilities for the tags, one might consider a more sophisticated approach such as that described in Kupiec (1992); one could also use methods that depend on a small hand-tagged seed corpus, as suggested by one reviewer.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Our \u201cProjection\u201d baseline is able to benefit from the bilingual information and greatly improves upon the monolingual baselines, but falls short of the \u201cNo LP\u201d model by 2.5% on an average.","label":0},{"source_text":"There is no global pruning.","target_text":"Sie.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Evaluation of the Segmentation as a Whole.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The likelihood of the observed data under the model is where P(yi, xi) is defined as in (9).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For each pair of judges consider one judge as the standard,.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"By making use of global context, it has achieved excellent results on both MUC6 and MUC7 official test data.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(Other classes handled by the current system are discussed in Section 5.)","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This smooth guarantees that there are no zeroes estimated.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"We experiment with four values for each hyperparameter resulting in 16 (\u00ce\u00b1, \u00ce\u00b2) combinations: \u00ce\u00b1 \u00ce\u00b2 0.001, 0.01, 0.1, 1.0 0.01, 0.1, 1.0, 10 Iterations In each run, we performed 30 iterations of Gibbs sampling for the type assignment variables W .4 We use the final sample for evaluation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The semantic agreement KS eliminates some candidates, but also provides positive evidence in one case: if the candidate and anaphor both have semantic tags human, company, date, or location that were assigned via NER or the manually labeled dictionary entries.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Asking the annotator to also formulate the question is a way of arriving at more reproducible decisions.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Intuitively, as suggested by the example in the introduction, this is the right granularity to capture domain effects.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Lexicalizing several POS tags improves performance.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"These are shown, with their associated costs, as follows: ABj nc 4.0 AB C\/jj 6.0 CD \/vb 5.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Using cn to denote the number of n-grams, total memory consumption of TRIE, in bits, is plus quantization tables, if used.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Many hanzi have more than one pronunciation, where the correct.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In the experiments below, we employ a data-driven deterministic dependency parser producing labeled projective dependency graphs,3 previously tested on Swedish (Nivre et al., 2004) and English (Nivre and Scholz, 2004).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"However, the point of RandLM is to scale to even larger data, compensating for this loss in quality.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For the experiments in this paper we made a couple of additional modifications to the CoBoost algorithm.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"49 99.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"na me =>2 ha nzi fa mi ly 1 ha nzi gi ve n 6.1 ha nzi fa mi ly => ha nz ii 7.2 ha nzi fa mi ly => ha nzi i ha nz ij 8.1 ha nzi gi ve n => ha nz ii 9.2 ha nzi giv en => ha nzi i ha nz ij The difficulty is that given names can consist, in principle, of any hanzi or pair of hanzi, so the possible given names are limited only by the total number of hanzi, though some hanzi are certainly far more likely than others.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Log-linear combination (loglin) improves on this in all cases, and also beats the pure IN system.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Once we figure out the important word (e.g. keyword), we believe we can capture the meaning of the phrase by the keyword.","label":0},{"source_text":"There is no global pruning.","target_text":"Additionally, for a given coverage set, at most 250 different hypotheses are kept during the search process, and the number of different words to be hypothesized by a source word is limited.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The last four columns in Table 3 show the distribution of nonprojective arcs with respect to the number of lifts required.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The second setting uses the news-related subcorpora for the NIST09 MT Chinese to English evaluation8 as IN, and the remaining NIST parallel Chinese\/English corpora (UN, Hong Kong Laws, and Hong Kong Hansard) as OUT.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Each trie node is individually allocated and full 64-bit pointers are used to find them, wasting memory.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"This pumping lemma states that if there is tree, t = t2t3t4t5, generated by a TAG G, such that its height is more than a predetermined bound k, then all trees of the form ti it tstt ts for each i > 0 will also generated by G. Similarly, for tree sets with independent paths and more complex path sets, tree pumping lemmas can be given.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Unique Occurrences and Zone (UNIQ): This group of features indicates whether the word is unique in the whole document.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(1991}, Gu and Mao (1994), and Nie, Jin, and Hannan (1994).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Their default variant implements a forward trie, in which words are looked up in their natural left-to-right order.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The PCFG was trained from the same sections of the Penn Treebank as the other three parsers.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The price shocked her.), or same-kind (e.g., Her health insurance paid for the hospital fees, but the automobile insurance did not cover the repair.).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Our code has been publicly available and intergrated into Moses since October 2010.","label":0},{"source_text":"Here we present two algorithms.","target_text":"It's not clear how to apply these methods in the unsupervised case, as they required cross-validation techniques: for this reason we use the simpler smoothing method shown here. input to the unsupervised algorithm is an initial, &quot;seed&quot; set of rules.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"It is a relatively frequent word in the domain, but it can be used in different extraction scenarios.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"However, reads in the TRIE data structure are more expensive due to bit-level packing, so we found that it is faster to use interpolation search the entire time.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"In Semitic languages the situation is very different.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"and f,.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"For all variants, we found that BerkeleyLM always rounds the floating-point mantissa to 12 bits then stores indices to unique rounded floats.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Unsupervised Part-of-Speech Tagging with Bilingual Graph-Based Projections","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For example: McCann initiated a new global system.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Using cn to denote the number of n-grams, total memory consumption of TRIE, in bits, is plus quantization tables, if used.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In a grammar which generates independent paths the derivations of sibling constituents can not share an unbounded amount of information.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Here we push the single-framework conjecture across the board and present a single model that performs morphological segmentation and syntactic disambiguation in a fully generative framework.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The percentage scores on the axis labels represent the amount of variation in the data explained by the dimension in question.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This is less effective in our setting, where IN and OUT are disparate.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Yet, some hanzi are far more probable in women's names than they are in men's names, and there is a similar list of male-oriented hanzi: mixing hanzi from these two lists is generally less likely than would be predicted by the independence model.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"AdaBoost finds a weighted combination of simple (weak) classifiers, where the weights are chosen to minimize a function that bounds the classification error on a set of training examples.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Twentieth-century linguistic work on Chinese (Chao 1968; Li and Thompson 1981; Tang 1988,1989, inter alia) has revealed the incorrectness of this traditional view.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Table 1 shows four words \u00e2\u0080\u009c 0 Indeed NN Indeed Saddamwhose unvocalized surface forms 0 an are indistinguishable.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Find keywords for each NE pair When we look at the contexts for each domain, we noticed that there is one or a few important words which indicate the relation between the NEs (for example, the word \u00e2\u0080\u009cunit\u00e2\u0080\u009d for the phrase \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Discriminative Instance Weighting for Domain Adaptation in Statistical Machine Translation","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This model admits a simple Gibbs sampling algorithm where the number of latent variables is proportional to the number of word types, rather than the size of a corpus as for a standard HMM sampler (Johnson, 2007).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The problem is to store these two values for a large and sparse set of n-grams in a way that makes queries efficient.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The number of ATB n-grams also falls below the WSJ sample size as the largest WSJ sample appeared in only 162 corpus positions.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We use the default inference parameters.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"AdaBoost.MH maintains a distribution over instances and labels; in addition, each weak-hypothesis outputs a confidence vector with one confidence value for each possible label.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"As a result, Habash et al.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The second stage links sets which involve the same pairs of individual NEs.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"They return a value in the range [0,1], where 0 indicates neutrality and 1 indicates the strongest belief that the candidate and anaphor are coreferent.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In this paper, we show how non-projective dependency parsing can be achieved by combining a datadriven projective parser with special graph transformation techniques.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For example, the two NEs \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d and \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d have the following contexts.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For a language like English, this problem is generally regarded as trivial since words are delimited in English text by whitespace or marks of punctuation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Ex: The government said it ...","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Morphological disambiguators that consider a token in context (an utterance) and propose the most likely morphological analysis of an utterance (including segmentation) were presented by Bar-Haim et al. (2005), Adler and Elhadad (2006), Shacham and Wintner (2007), and achieved good results (the best segmentation result so far is around 98%).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Then each arc of D maps either from an element of H to an element of p, or from E-i.e., the empty string-to an element of P. More specifically, each word is represented in the dictionary as a sequence of arcs, starting from the initial state of D and labeled with an element 5 of Hxp, which is terminated with a weighted arc labeled with an element of Ex P. The weight represents the estimated cost (negative log probability) of the word.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Alternatively, h can be thought of as defining a decision list of rules x y ranked by their &quot;strength&quot; h(x, y).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We aligned the texts at a sentence level across all four languages, resulting in 1064 sentence per language.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"IRSTLM and BerkeleyLM use this state function (and a limit of N \u22121 words), but it is more strict than necessary, so decoders using these packages will miss some recombination opportunities.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Note that these observa sider suffix features, capitalization features, punctuation, and digit features.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Except our own and MENE + reference resolution, the results in Table 6 are all official MUC7 results.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We attempt to formalize this notion in terms of the tee pumping lemma which can be used to show that a tee set does not have dependent paths.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"72 78.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The extent to which this constraint is enforced varies greatly across existing methods.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"In the namedentity problem each example is a (spelling,context) pair.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Mohri [1995]) shows promise for improving this situation.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"RandLM 0.2 (Talbot and Osborne, 2007) stores large-scale models in less memory using randomized data structures.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"While RST (Mann, Thompson 1988) proposed that a single relation hold between adjacent text segments, SDRT (Asher, Lascarides 2003) maintains that multiple relations may hold simultaneously.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"1 61.2 43.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"3 60.7 50.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We identified three ways that contextual roles can be exploited: (1) by identifying caseframes that co-occur in resolutions, (2) by identifying nouns that co-occur with case- frames and using them to crosscheck anaphor\/candidate compatibility, (3) by identifying semantic classes that co- occur with caseframes and using them to crosscheck anaphor\/candidate compatability.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The bias of automatic methods in favor of statistical systems seems to be less pronounced on out-of-domain test data.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"A very small excerpt from an Italian-English graph is shown in Figure 1.","label":0},{"source_text":"Here we present two algorithms.","target_text":"AdaBoost.MH maintains a distribution over instances and labels; in addition, each weak-hypothesis outputs a confidence vector with one confidence value for each possible label.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"A sentence was withheld from section 22 because its extreme length was troublesome for a couple of the parsers.'","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"2.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Statistics are printed before Moses exits and after parts of the decoder have been destroyed.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"This feature ft incorporates information from the smoothed graph and prunes hidden states that are inconsistent with the thresholded vector tx.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"All commentaries have been tagged with part-of-speech information using Brants\u00e2\u0080\u0099 TnT1 tagger and the Stuttgart\/Tu\u00c2\u00a8bingen Tag Set automatic analysis was responsible for this decision.)","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Thus at each iteration the method induces at most n x k rules, where k is the number of possible labels (k = 3 in the experiments in this paper). step 3.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"36.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Moving beyond directly related work, major themes in SMT adaptation include the IR (Hildebrand et al., 2005; L\u00a8u et al., 2007; Zhao et al., 2004) and mixture (Finch and Sumita, 2008; Foster and Kuhn, 2007; Koehn and Schroeder, 2007; L\u00a8u et al., 2007) approaches for LMs and TMs described above, as well as methods for exploiting monolingual in-domain text, typically by translating it automatically and then performing self training (Bertoldi and Federico, 2009; Ueffing et al., 2007; Schwenk and Senellart, 2009).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We also collapse unary chains withidentical basic categories like NP \u00e2\u0086\u0092 NP.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"Memory-based classifiers for the experiments were created using TiMBL (Daelemans et al., 2003).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"A final alternate approach would be to combine weighted joint frequencies rather than conditional estimates, ie: cI(s, t) + w,\\(s, t)co(, s, t), suitably normalized.5 Such an approach could be simulated by a MAP-style combination in which separate 0(t) values were maintained for each t. This would make the model more powerful, but at the cost of having to learn to downweight OUT separately for each t, which we suspect would require more training data for reliable performance.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Other packages walk their respective data structures once to find wnf and again to find {b(wn\u22121 i )}f\u22121 i=1if necessary.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Again, the monotone search performs worst.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"4.2 Global Features.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In Section 4, we present the performance measures used and give translation results on the Verbmobil task.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"This paper, however, provides a comprehensive overview of the data collection effort and its current state.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We have shown the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Each parse is converted into a set of constituents represented as a tuples: (label, start, end).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Our suspicion is that BLEU is very sensitive to jargon, to selecting exactly the right words, and not synonyms that human judges may appreciate as equally good.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"MCTAG's are able to generate tee sets having dependent paths.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Collisions between two keys in the table can be identified at model building time.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"). context=x The context for the entity.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The Potsdam Commentary Corpus","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Decreasing the threshold results in higher mWER due to additional search errors.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"To combat the proliferation of parsing edges, we prune the lattices according to a hand-constructed lexicon of 31 clitics listed in the ATB annotation guidelines (Maamouri et al., 2009a).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Hieu Hoang named the code \u201cKenLM\u201d and assisted with Moses along with Barry Haddow.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"(Thus the domain of the dev and test corpora matches IN.)","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"This overview is illustrated in Figure 1.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Supervised part-of-speech (POS) taggers, for example, approach the level of inter-annotator agreement (Shen et al., 2007, 97.3% accuracy for English).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The test data was again drawn from a segment of the Europarl corpus from the fourth quarter of 2000, which is excluded from the training data.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Whereas Arabic linguistic theory as Saddam (a) Reference (b) Stanford signs (1) and (2) to the class of pseudo verbs 01 +i J>1\u00ef\u00bf\u00bd inna and her sisters since they can beinflected, the ATB conventions treat (2) as a com plementizer, which means that it must be the head of SBAR.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"One is that smaller sets sometime have meaningless keywords, like \u00e2\u0080\u009cstrength\u00e2\u0080\u009d or \u00e2\u0080\u009cadd\u00e2\u0080\u009d in the CC-domain, or \u00e2\u0080\u009ccompare\u00e2\u0080\u009d in the PC-domain.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"VBD she added VP PUNC \u00e2\u0080\u009c SBAR IN NP 0 NN.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Consequently, we implemented our own annotation tool ConAno in Java (Stede, Heintze 2004), which provides specifically the functionality needed for our purpose.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A related point is that mutual information is helpful in augmenting existing electronic dictionaries, (cf.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Even without features, but still using the tag prior, our median result is 52.0%, still significantly outperforming Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"4.3 Translation Experiments.","label":0},{"source_text":"There is no global pruning.","target_text":"For \u00c3\u0086 = 1, a new target language word is generated using the trigram language model p(eje0; e00).","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We pick the parse that is most similar to the other parses by choosing the one with the highest sum of pairwise similarities.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"As can be seen, GR and this \"pared-down\" statistical method perform quite similarly, though the statistical method is still slightly better.16 AG clearly performs much less like humans than these methods, whereas the full statistical algorithm, including morphological derivatives and names, performs most closely to humans among the automatic methods.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"This is the first set that gives us a fair evaluation of the Bayes models, and the Bayes switching model performs significantly better than its non-parametric counterpart.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Figure 1 reveals that an event that \u00e2\u0080\u009cdamaged\u00e2\u0080\u009d objects may also cause injuries; a disaster that \u00e2\u0080\u009coccurred\u00e2\u0080\u009d may be investigated to find its \u00e2\u0080\u009ccause\u00e2\u0080\u009d; a disaster may \u00e2\u0080\u009cwreak\u00e2\u0080\u009d havoc as it \u00e2\u0080\u009ccrosses\u00e2\u0080\u009d geographic regions; and vehicles that have a \u00e2\u0080\u009cdriver\u00e2\u0080\u009d may also \u00e2\u0080\u009ccarry\u00e2\u0080\u009d items.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Still, for about good number of sentences, we do have this direct comparison, which allows us to apply the sign test, as described in Section 2.2.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Our system does not currently make use of titles, but it would be straightforward to do so within the finite-state framework that we propose.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Beyond optimizing the memory size of TRIE, there are alternative data structures such as those in Guthrie and Hepple (2010).","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"At first glance, the problem seems quite complex: a large number of rules is needed to cover the domain, suggesting that a large number of labeled examples is required to train an accurate classifier.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Almost all annotators expressed their preference to move to a ranking-based evaluation in the future.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Otherwise, the scope of the search problem shrinks recursively: if A[pivot] < k then this becomes the new lower bound: l +\u2014 pivot; if A[pivot] > k then u +\u2014 pivot.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Although these are technically nominal, they have become known as \u00e2\u0080\u009cequational\u00e2\u0080\u009d sentences.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The use of ILP in learning the desired grammar significantly increases the computational complexity of this method.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We have checked if there are similar verbs in other major domains, but this was the only one.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Both parameters depend on a single hyperparameter \u00ce\u00b1.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The ATB is disadvantaged by having fewer trees with longer average 5 LDC A-E catalog numbers: LDC2008E61 (ATBp1v4), LDC2008E62 (ATBp2v3), and LDC2008E22 (ATBp3v3.1).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We paid particular attention to minimize the number of free parameters, and used the same hyperparameters for all language pairs, rather than attempting language-specific tuning.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This suggests a direct parallel to (1): where \u02dcp(s, t) is a joint empirical distribution extracted from the IN dev set using the standard procedure.2 An alternative form of linear combination is a maximum a posteriori (MAP) combination (Bacchiani et al., 2004).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Two common cases are the attribu tive adjective and the process nominal _; maSdar, which can have a verbal reading.4 At tributive adjectives are hard because they are or- thographically identical to nominals; they are inflected for gender, number, case, and definiteness.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The Stanford parser includes both the manually annotated grammar (\u00c2\u00a74) and an Arabic unknown word model with the following lexical features: 1.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Evaluation within a set The evaluation of paraphrases within a set of phrases which share a keyword is illustrated in Figure 4.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"\u2014 I would also like to point out to commissioner Liikanen that it is not easy to take a matter to a national court.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Find keywords for each NE pair The keywords are found for each NE category pair.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"3.2 Reordering with IBM Style.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In a more recent study than Chang et al., Wang, Li, and Chang (1992) propose a surname-driven, non-stochastic, rule-based system for identifying personal names.17 Wang, Li, and Chang also compare their performance with Chang et al.'s system.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Central to our approach (see Algorithm 1) is a bilingual similarity graph built from a sentence-aligned parallel corpus.","label":0},{"source_text":"There is no global pruning.","target_text":"Search CPU time mWER SSER Method [sec] [%] [%] MonS 0:9 42:0 30:5 QmS 10:6 34:4 23:8 IbmS 28:6 38:2 26:2 4.2 Performance Measures.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"We showed in \u00c2\u00a72 that lexical ambiguity explains the underperformance of these categories.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Overview of the method 2.2 Step by Step Algorithm.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"These two principles guide experimentation in this framework, and together with the evaluation measures help us decide which specific type of substructure to combine.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"As in the case of the derivation trees of CFG's, nodes are labeled by a member of some finite set of symbols (perhaps only implicit in the grammar as in TAG's) used to denote derived structures.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We map the ATB morphological analyses to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags for all experiments.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The average agreement among the human judges is .76, and the average agreement between ST and the humans is .75, or about 99% of the interhuman agreement.15 One can better visualize the precision-recall similarity matrix by producing from that matrix a distance matrix, computing a classical metric multidimensional scaling (Torgerson 1958; Becker, Chambers, Wilks 1988) on that dis\u00c2\u00ad tance matrix, and plotting the first two most significant dimensions.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Each round is composed of two stages; each stage updates one of the classifiers while keeping the other classifier fixed.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"They also describe an application of cotraining to classifying web pages (the to feature sets are the words on the page, and other pages pointing to the page).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"MENE (Maximum Entropy Named Entity) (Borth- wick, 1999) was combined with Proteus (a hand- coded system), and came in fourth among all MUC 7 participants.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Its only purpose is 3 This follows since each \u00ce\u00b8t has St \u00e2\u0088\u0092 1 parameters and.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"For our demonstration system, we typically use the pruning threshold t0 = 5:0 to speed up the search by a factor 5 while allowing for a small degradation in translation accuracy.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The Levenshtein distance between the automatic translation and each of the reference translations is computed, and the minimum Levenshtein distance is taken.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"In this section, we will explain the algorithm step by step with examples.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Training and testing is based on the Europarl corpus.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"We are given a source string fJ 1 = f1:::fj :::fJ of length J, which is to be translated into a target string eI 1 = e1:::ei:::eI of length I. Among all possible target strings, we will choose the string with the highest probability: ^eI 1 = arg max eI 1 fPr(eI 1jfJ 1 )g = arg max eI 1 fPr(eI 1) Pr(fJ 1 jeI 1)g : (1) The argmax operation denotes the search problem, i.e. the generation of the output sentence in the target language.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Two issues distinguish the various proposals.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We use w erations of sampling (see Figure 2 for a depiction).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Therefore, we want state to encode the minimum amount of information necessary to properly compute language model scores, so that the decoder will be faster and make fewer search errors.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We extend the Matsoukas et al approach in several ways.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"7 68.3 56.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We thank Felix Hageloh (Hageloh, 2006) for providing us with this version. proposed in (Tsarfaty, 2006).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Topicalization of NP subjects in SVO configurations causes confusion with VO (pro-drop).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For example, hanzi containing the INSECT radical !R tend to denote insects and other crawling animals; examples include tr wal 'frog,' feng1 'wasp,' and !Itt she2 'snake.'","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The first author is supported by a National Defense Science and Engineering Graduate (NDSEG) fellowship.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For the PROBING implementation, hash table sizes are in the millions, so the most relevant values are on the right size of the graph, where linear probing wins.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The paper is organized as follows: Section 2 explains the different layers of annotation that have been produced or are being produced.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"For RandLM and IRSTLM, the effect of caching can be seen on speed and memory usage.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"2.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"We show how a datadriven deterministic dependency parser, in itself restricted to projective structures, can be combined with graph transformation techniques to produce non-projective structures.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"When the maSdar lacks a determiner, the constituent as a whole resem bles the ubiquitous annexation construct \u00ef\u00bf\u00bd ?f iDafa.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"So, who won the competition?","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Our annotators pointed out that very often they made almost random decisions as to what relation to choose, and where to locate the boundary of a span.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We obtained positive results using a very simple phrase-based system in two different adaptation settings: using English\/French Europarl to improve a performance on a small, specialized medical domain; and using non-news portions of the NIST09 training material to improve performance on the news-related corpora.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"x EG, has agreed to be bought by H x EG, now owned by H x H to acquire EG x H\u00e2\u0080\u0099s agreement to buy EG Three of those phrases are actually paraphrases, but sometime there could be some noise; such as the second phrase above.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"We therefore used the arithmetic mean of each interjudge precision-recall pair as a single measure of interjudge similarity.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"(2010)\u00e2\u0080\u0099s richest model: optimized via either EM or LBFGS, as their relative performance depends on the language.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The normalization factor plays an important role in the AdaBoost algorithm.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The natural baseline (baseline) outperforms the pure IN system only for EMEA\/EP fren.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"summit Sharm (a) Al-Sheikh summit Sharm (b) DTNNP Al-Sheikh in a corpus position without a bracketing label, then we also add \u00e2\u0088\u0097n, NIL) to M. We call the set of unique n-grams with multiple labels in M the variation nuclei of C. Bracketing variation can result from either annotation errors or linguistic ambiguity.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Hence we decided to select ten commentaries to form a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099, for which the entire range of annotation levels was realized, so that experiments with multi-level querying could commence.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This WFST is then summed with the WFST implementing the dictionary and morphological rules, and the transitive closure of the resulting transducer is computed; see Pereira, Riley, and Sproat (1994) for an explanation of the notion of summing WFSTs.12 Conceptual Improvements over Chang et al.'s Model.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"When the same token is to be interpreted as a single lexeme fmnh, it may function as a single adjective \u201cfat\u201d.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We extend Subramanya et al.\u2019s intuitions to our bilingual setup.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Another way to interpret this is that less than 5% of the correct constituents are missing from the hypotheses generated by the union of the three parsers.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"It uses a maximum entropy framework and classifies each word given its features.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Graph construction does not require any labeled data, but makes use of two similarity functions.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The details are given in (Och and Ney, 2000).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"A more forceful approach for encoding sparsity is posterior regularization, which constrains the posterior to have a small number of expected tag assignments (Grac\u00c2\u00b8a et al., 2009).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(See also Wu and Fung [1994].)","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"hanzi in the various name positions, derived from a million names.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This is to allow for fair comparison between the statistical method and GR, which is also purely dictionary-based.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The model starts by generating a tag assignment for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"0 70.9 42.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Our System Wang, Li, and Chang a. 1\\!f!IP Eflltii \/1\\!f!J:P $1til I b. agm: I a m: c. 5 Bf is Bf 1 d. \"*:t: w _t ff 1 \"* :t: w_tff 1 g., , Transliteration\/Translation chen2zhongl-shenl qu3 'music by Chen Zhongshen ' huang2rong2 youlyoul de dao4 'Huang Rong said soberly' zhangl qun2 Zhang Qun xian4zhang3 you2qingl shang4ren2 hou4 'after the county president You Qing had assumed the position' lin2 quan2 'Lin Quan' wang2jian4 'Wang Jian' oulyang2-ke4 'Ouyang Ke' yinl qi2 bu4 ke2neng2 rong2xu3 tai2du2 er2 'because it cannot permit Taiwan Independence so' silfa3-yuan4zhang3 lin2yang2-gang3 'president of the Judicial Yuan, Lin Yanggang' lin2zhangl-hu2 jiangl zuo4 xian4chang3 jie3shuol 'Lin Zhanghu will give an ex\u00c2\u00ad planation live' jin4\/iang3 nian2 nei4 sa3 xia4 de jinlqian2 hui4 ting2zhi3 'in two years the distributed money will stop' gaoltangl da4chi2 ye1zi0 fen3 'chicken stock, a tablespoon of coconut flakes' you2qingl ru4zhu3 xian4fu3 lwu4 'after You Qing headed the county government' Table 5 Performance on morphological analysis.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We model each parse as the decisions made to create it, and model those decisions as independent events.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"As in boosting, the algorithm works in rounds.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Hyperparameter settings are sorted according to the median one-to-one metric over runs.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Location list is processed into a list of unigrams and bigrams (e.g., New York).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Reflexive pronouns with only 1 NP in scope..","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"First, we use a novel graph-based framework for projecting syntactic information across language boundaries.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"2.1.1 Lexical Seeding It is generally not safe to assume that multiple occurrences of a noun phrase refer to the same entity.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This leads to a linear combination of domain-specific probabilities, with weights in [0, 1], normalized to sum to 1.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Here we push the single-framework conjecture across the board and present a single model that performs morphological segmentation and syntactic disambiguation in a fully generative framework.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"To control for the effect of the HSPELL-based pruning, we also experimented with a morphological analyzer that does not perform this pruning.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, in MUC6, there are four zones (TXT, HL, DATELINE, DD).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We would like to thank Prof. Ralph Grish- man, Mr. Takaaki Hasegawa and Mr. Yusuke Shinyama for useful comments, discussion and evaluation.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"This work was funded by NSF grant IRI-9502312.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Table 1 The cost as a novel given name (second position) for hanzi from various radical classes.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The table shows that the lexicon tag frequency predicated by our full model are the closest to the gold standard.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Finally, we add \u00e2\u0080\u009cDT\u00e2\u0080\u009d to the tags for definite nouns and adjectives (Kulick et al., 2006).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This paper describes the several performance techniques used and presents benchmarks against alternative implementations.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Sentence pairs are the natural instances for SMT, but sentences often contain a mix of domain-specific and general language.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"(2010), we adopt a simpler na\u00c2\u00a8\u00c4\u00b1ve Bayes strategy, where all features are emitted independently.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The rhetorical structure annotations of PCC have all been converted to URML.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The final strong hypothesis, denoted 1(x), is then the sign of a weighted sum of the weak hypotheses, 1(x) = sign (Vii atht(x)), where the weights at are determined during the run of the algorithm, as we describe below.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Hieu Hoang named the code \u201cKenLM\u201d and assisted with Moses along with Barry Haddow.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"5.2 Discussion.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The SynRole KS computes the relative frequency with which the candidates\u00e2\u0080\u0099 syntactic role (subject, direct object, PP object) appeared in resolutions in the training set.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"A similar maximumlikelihood approach was used by Foster and Kuhn (2007), but for language models only.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Specifically, we (+FEATS) utilizes the tag prior as well as features (e.g., suffixes and orthographic features), discussed in Section 3, for the P (W |T , \u00cf\u0088) component.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Because we don\u2019t have a separate development set, we used the training set to select among them and found 0.2 to work slightly better than 0.1 and 0.3.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Taking only the highest frequency rules is much &quot;safer&quot;, as they tend to be very accurate.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Similar advances have been made in machine translation (Frederking and Nirenburg, 1994), speech recognition (Fiscus, 1997) and named entity recognition (Borthwick et al., 1998).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"(7) \u03c6 s,t This is a somewhat less direct objective than used by Matsoukas et al, who make an iterative approximation to expected TER.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"This allow the learners to &quot;bootstrap&quot; each other by filling the labels of the instances on which the other side has abstained so far.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For instance, for TTS it is necessary to know that a particular sequence of hanzi is of a particular category because that knowl\u00c2\u00ad edge could affect the pronunciation; consider, for example the issues surrounding the pronunciation of ganl I qian2 discussed in Section 1.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"3.1 Lexicon Component.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"A promising direction for future work is to explicitly model a distribution over tags for each word type.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Fortunately, performance was stable across various values, and we were able to use the same hyperparameters for all languages.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"3 58.3 40.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"\u2022 Similarly, when the na\u00efve Bayes classifier is configured such that the constituents require estimated probabilities strictly larger than 0.5 to be accepted, there is not enough probability mass remaining on crossing brackets for them to be included in the hypothesis.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Figure 3 shows a small fragment of the WFST encoding the dictionary, containing both entries forjust discussed, g:t\u00c2\u00a5 zhonglhua2 min2guo2 (China Republic) 'Republic of China,' and i\u00c2\u00a5inl.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Foreign names are usually transliterated using hanzi whose sequential pronunciation mimics the source language pronunciation of the name.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The main disadvantage of manual evaluation is that it is time-consuming and thus too expensive to do frequently.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"When keys are longer than 64 bits, we conserve space by replacing the keys with their 64-bit hashes.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This model is easily incorporated into the segmenter by building a WFST restrict\u00c2\u00ad ing the names to the four licit types, with costs on the arcs for any particular name summing to an estimate of the cost of that name.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The translation search is carried out with the category markers and the city names are resubstituted into the target sentence as a postprocessing step.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Each model was able to produce hypotheses for all input sentences.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"While we have minimized forward-looking state in Section 4.1, machine translation systems could also benefit by minimizing backward-looking state.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Diacritics can also be used to specify grammatical relations such as case and gender.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"This paper is based on work supported in part by DARPA through IBM.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Find keywords for each NE pair The keywords are found for each NE category pair.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Syntactic decoders, such as cdec (Dyer et al., 2010), build state from null context then store it in the hypergraph node for later extension.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Tsarfaty (2006) argues that for Semitic languages determining the correct morphological segmentation is dependent on syntactic context and shows that increasing information sharing between the morphological and the syntactic components leads to improved performance on the joint task.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Juri Ganitkevitch answered questions about Joshua.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"However, in practice, unknown word models also make the distribution improper.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The combining algorithm is presented with the candidate parses and asked to choose which one is best.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"We suggest that in unlexicalized PCFGs the syntactic context may be explicitly modeled in the derivation probabilities.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"While the Bootstrap method is slightly more sensitive, it is very much in line with the sign test on text blocks.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Step 3.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For example, take ..., says Maury Cooper, a vice president at S.&P.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"As indicated by bolding, for seven out of eight languages the improvements of the \u201cWith LP\u201d setting are statistically significant with respect to the other models, including the \u201cNo LP\u201d setting.11 Overall, it performs 10.4% better than the hitherto state-of-the-art feature-HMM baseline, and 4.6% better than direct projection, when we macro-average the accuracy over all languages.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"During development, we sensed that the Recency and Syn- role KSs did not deserve to be on equal footing with the other KSs because their knowledge was so general.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"of Articles No.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"This suggests that different types of anaphora may warrant different treatment: definite NP resolution may depend more on lexical semantics, while pronoun resolution may depend more on contextual semantics.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"So far the graph has been completely unlabeled.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We compare three hash tables: our probing implementation, GCC\u2019s hash set, and Boost\u2019s8 unordered.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Hearst 92) describes a method for extracting hyponyms from a corpus (pairs of words in &quot;isa&quot; relations).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"To set \u03b2, we used the same criterion as for \u03b1, over a dev corpus: The MAP combination was used for TM probabilities only, in part due to a technical difficulty in formulating coherent counts when using standard LM smoothing techniques (Kneser and Ney, 1995).3 Motivated by information retrieval, a number of approaches choose \u201crelevant\u201d sentence pairs from OUT by matching individual source sentences from IN (Hildebrand et al., 2005; L\u00a8u et al., 2007), or individual target hypotheses (Zhao et al., 2004).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Our annotators pointed out that very often they made almost random decisions as to what relation to choose, and where to locate the boundary of a span.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"ments contained 322 anaphoric links.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Like the string languages of MCTAG's, the complexity of the path set increases as the cardinality of the elementary tee sets increases, though both the string languages and path sets will always be semilinear.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Taking the intersection of languages in these resources, and selecting languages with large amounts of parallel data, yields the following set of eight Indo-European languages: Danish, Dutch, German, Greek, Italian, Portuguese, Spanish and Swedish.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"If the same pair of NE instances is used with different phrases, these phrases are likely to be paraphrases.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"One obvious application is information extraction.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Both parametric and non-parametric models are explored.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"All improvements over the baseline are statistically significant beyond the 0.01 level (McNemar\u2019s test).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"9 www.ling.unipotsdam.de\/sfb\/ Figure 2: Screenshot of Annis Linguistic Database 3.3 Symbolic and knowledge-based.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"phrase (markContainsVerb).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"If the key distribution\u2019s range is also known (i.e. vocabulary identifiers range from 0 to the number of words), then interpolation search can use this information instead of reading A[0] and A[|A |\u2212 1] to estimate pivots; this optimization alone led to a 24% speed improvement.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Microsoft\u2019s approach uses dependency trees, others use hierarchical phrase models.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"While many systems had similar performance, the results offer interesting insights, especially about the relative performance of statistical and rule-based systems.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"A corpus of German newspaper commentaries has been assembled at Potsdam University, and annotated with different linguistic information, to different degrees.","label":1},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Even if an example like this is not labeled, it can be interpreted as a &quot;hint&quot; that Mr and president imply the same category.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Although the ultimate goal of the Verbmobil project is the translation of spoken language, the input used for the translation experiments reported on in this paper is the (more or less) correct orthographic transcription of the spoken sentences.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In this situation, BABAR takes the conservative approach and declines to make a resolution.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Syntactic universals are a well studied concept in linguistics (Carnie, 2002; Newmeyer, 2005), and were recently used in similar form by Naseem et al. (2010) for multilingual grammar induction.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"For all languages we do not make use of a tagging dictionary.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Lossy compressed models RandLM (Talbot and Osborne, 2007) and Sheffield (Guthrie and Hepple, 2010) offer better memory consumption at the expense of CPU and accuracy.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"In this paper, Section 2 begins by explaining how contextual role knowledge is represented and learned.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Third, we develop a human interpretable grammar that is competitive with a latent variable PCFG.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We hypothesize that modeling morphological information will greatly constrain the set of possible tags, thereby further refining the representation of the tag lexicon.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The baseline alignment model does not permit that a source word is aligned to two or more target words, e.g. for the translation direction from German toEnglish, the German compound noun 'Zahnarztter min' causes problems, because it must be translated by the two target words dentist's appointment.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In the terrorism domain, 1600 texts were used for training and the 40 test docu X \u00e2\u0088\u00a9Y =\u00e2\u0088\u0085 All sets of hypotheses (and their corresponding belief values) in the current model are crossed with the sets of hypotheses (and belief values) provided by the new evidence.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Combining multiple highly-accurate independent parsers yields promising results.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"These 140 NE categories are designed by extending MUC\u00e2\u0080\u0099s 7 NE categories with finer sub-categories (such as Company, Institute, and Political Party for Organization; and Country, Province, and City for Location) and adding some new types of NE categories (Position Title, Product, Event, and Natural Object).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"5.2 Setup.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(4) is minimized by setting Since a feature may be present in only a few examples, W_ can be in practice very small or even 0, leading to extreme confidence values.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"keywords Step 3 Sets of phrases based on keywords Step 4 Links between sets of phrases All the contexts collected for a given domain are gathered in a bag and the TF\/ITF scores are calculated for all the words except stopwords in the bag.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"For irt the Good-Turing estimate just discussed gives us an estimate of p(unseen(f,) I f,)-the probability of observing a previously unseen instance of a construction in ft given that we know that we have a construction in f,.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Although we have not done any experiments on other languages, this way of using global features from a whole document should be applicable to other languages.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Blum and Mitchell 98) give an example that illustrates just how powerful the second constraint can be.","label":0},{"source_text":"There is no global pruning.","target_text":"For the translation model Pr(fJ 1 jeI 1), we go on the assumption that each source word is aligned to exactly one target word.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Typically, judges initially spent about 3 minutes per sentence, but then accelerate with experience.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"For comparison, the completely unsupervised feature-HMM baseline accuracy on the universal POS tags for English is 79.4%, and goes up to 88.7% with a treebank dictionary.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We propose maximizing , where is the sequence of named- entity tags assigned to the words in the sentence , and is the information that can be extracted from the whole document containing . Our system is built on a maximum entropy classifier.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"For example, in .., says Mr. Cooper, a vice president of.. both a spelling feature (that the string contains Mr.) and a contextual feature (that president modifies the string) are strong indications that Mr. Cooper is of type Person.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"TRIE uses less memory and has better locality.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Schapire and Singer show that the training error is bounded above by Thus, in order to greedily minimize an upper bound on training error, on each iteration we should search for the weak hypothesis ht and the weight at that minimize Z.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Before explaining our method in detail, we present a brief overview in this subsection.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"If one of these checks fails then this knowledge source reports that the candidate is not a viable antecedent for the anaphor.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"One obvious application is information extraction.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In this way, the method reported on here will necessarily be similar to a greedy method, though of course not identical.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"21 In Chinese, numerals and demonstratives cannot modify nouns directly, and must be accompanied by.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Note that on some examples (around 2% of the test set) CoBoost abstained altogether; in these cases we labeled the test example with the baseline, organization, label.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Section 4 draws some conclusions from the present state of the effort.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In a few cases, the criteria for correctness are made more explicit.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Extract NE instance pairs with contexts First, we extract NE pair instances with their context from the corpus.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"A corpus of German newspaper commentaries has been assembled and annotated with different information (and currently, to different degrees): part-of-speech, syntax, rhetorical structure, connectives, co-reference, and information structure.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The developers aimed to reduce memory consumption at the expense of time.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Table 2 shows our complete set of results.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Ties are rare in Bayes switching because the models are fine-grained \u2014 many estimated probabilities are involved in each decision.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"10 Other orthographic normalization schemes have been suggested for Arabic (Habash and Sadat, 2006), but we observe negligible parsing performance differences between these and the simple scheme used in this evaluation.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"While possible to utilize the feature-based log-linear approach described in Berg-Kirkpatrick et al.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We observe similar trends when using another measure \u00e2\u0080\u0093 type-level accuracy (defined as the fraction of words correctly assigned their majority tag), according to which La ng ua ge M etr ic B K 10 E M B K 10 L B F G S G 10 F EA T S B es t F EA T S M ed ia n E ng lis h 1 1 m 1 4 8 . 3 6 8 . 1 5 6 . 0 7 5 . 5 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 0 . 9 6 6 . 4 4 7 . 8 6 6 . 4 D an is h 1 1 m 1 4 2 . 3 6 6 . 7 4 2 . 6 5 8 . 0 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 2 . 1 6 1 . 2 4 3 . 2 6 0 . 7 D ut ch 1 1 m 1 5 3 . 7 6 7 . 0 5 5 . 1 6 4 . 7 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 6 . 4 6 9 . 0 5 1 . 5 6 7 . 3 Po rtu gu es e 1 1 m 1 5 0 . 8 7 5 . 3 4 3 . 2 7 4 . 8 44 .5 69 .2 6 4 . 1 7 4 . 5 5 6 . 5 7 0 . 1 S pa ni sh 1 1 m 1 \u00e2\u0080\u0093 \u00e2\u0080\u0093 4 0 . 6 7 3 . 2 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 8 . 3 6 8 . 9 5 0 . 0 5 7 . 2 Table 4: Comparison of our method (FEATS) to state-of-the-art methods.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"12 One class of full personal names that this characterization does not cover are married women's names.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Several extensions of AdaBoost for multiclass problems have been suggested (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The observed performance gains, coupled with the simplicity of model implementation, makes it a compelling alternative to existing more complex counterparts.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The method halves the error rate in comparison to a method using the labeled examples alone.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"In our case multi-threading is trivial because our data structures are read-only and uncached.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"2.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Thus, if one wants to segment words-for any purpose-from Chinese sentences, one faces a more difficult task than one does in English since one cannot use spacing as a guide.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Historically, Arabic grammar has identified two sentences types: those that begin with a nominal (\u00ef\u00bf\u00bd '.i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Formally, let el (62) be the number of classification errors of the first (second) learner on the training data, and let Eco be the number of unlabeled examples on which the two classifiers disagree.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE (without the help of hand-coded systems) has been shown to be somewhat inferior in performance.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Figure 2 shows examples of extracted NE pair instances and their contexts.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"This is the first set that gives us a fair evaluation of the Bayes models, and the Bayes switching model performs significantly better than its non-parametric counterpart.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Tokens tagged as PUNC are not discarded unless they consist entirely of punctuation.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This is motivated by taking \u03b2 po(s|t) to be the parameters of a Dirichlet prior on phrase probabilities, then maximizing posterior estimates p(s|t) given the IN corpus.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Many morphological decisions are based on long distance dependencies, and when the global syntactic evidence disagrees with evidence based on local linear context, the two models compete with one another, despite the fact that the PCFG takes also local context into account.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In this paper, we show how non-projective dependency parsing can be achieved by combining a datadriven projective parser with special graph transformation techniques.","label":1},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Now assume we have n pairs (xi,, x2,i) drawn from X1 X X2, where the first m pairs have labels whereas for i = m+ 1...n the pairs are unlabeled.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The number of ATB n-grams also falls below the WSJ sample size as the largest WSJ sample appeared in only 162 corpus positions.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"An important reason for separating the two types of features is that this opens up the possibility of theoretical analysis of the use of unlabeled examples.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This approach is similar to BABAR in that they both acquire knowledge from earlier resolutions.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"na me =>1 ha nzi fa mi ly 2 ha nzi gi ve n 3.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"1 61.2 43.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Following the system devised under the Qing emperor Kang Xi, hanzi have traditionally been classified according to a set of approximately 200 semantic radicals; members of a radical class share a particular structural component, and often also share a common meaning (hence the term 'semantic').","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The first point we need to address is what type of linguistic object a hanzi repre\u00c2\u00ad sents.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"A Stochastic Finite-State Word-Segmentation Algorithm for Chinese","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"This is the only important case, because otherwise the simple majority combining technique would pick the correct constituent.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Finally, the focus\/background partition is annotated, together with the focus question that elicits the corresponding answer.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The natural baseline (baseline) outperforms the pure IN system only for EMEA\/EP fren.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"If the key distribution\u2019s range is also known (i.e. vocabulary identifiers range from 0 to the number of words), then interpolation search can use this information instead of reading A[0] and A[|A |\u2212 1] to estimate pivots; this optimization alone led to a 24% speed improvement.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"3.1 Corpora.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"A position is presented by the word at that position.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The input for the segmentation task is however highly ambiguous for Semitic languages, and surface forms (tokens) may admit multiple possible analyses as in (BarHaim et al., 2007; Adler and Elhadad, 2006).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The combining algorithm is presented with the candidate parses and asked to choose which one is best.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Unsupervised Part-of-Speech Tagging with Bilingual Graph-Based Projections","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We have argued that the proposed method performs well.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"For each sentence, we counted how many n-grams in the system output also occurred in the reference translation.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"In both cases, SRILM walks its trie an additional time to minimize context as mentioned in Section 4.1.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The Stanford parser includes both the manually annotated grammar (\u00c2\u00a74) and an Arabic unknown word model with the following lexical features: 1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Furthermore, even the size of the dictionary per se is less important than the appropriateness of the lexicon to a particular test corpus: as Fung and Wu (1994) have shown, one can obtain substantially better segmentation by tailoring the lexicon to the corpus to be segmented.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We simulate lexical constraints by using an external lexical resource against which we verify whether OOV segments are in fact valid Hebrew lexemes.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"In the interest of testing the robustness of these combining techniques, we added a fourth, simple nonlexicalized PCFG parser.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Section 4 draws some conclusions from the present state of the effort.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Across all languages, +PRIOR consistently outperforms 1TW, reducing error on average by 9.1% and 5.9% on best and median settings respectively.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"TRIE uses less memory and has better locality.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"Even when there is training data available in the domain of interest, there is often additional data from other domains that could in principle be used to improve performance.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"This is by now a fairly standard representation for multiple morphological segmentation of Hebrew utterances (Adler, 2001; Bar-Haim et al., 2005; Smith et al., 2005; Cohen and Smith, 2007; Adler, 2007).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"SRILM (Stolcke, 2002) is widely used within academia.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Applications such as machine translation use language model probability as a feature to assist in choosing between hypotheses.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"To check for this, we do pairwise bootstrap resampling: Again, we repeatedly sample sets of sentences, this time from both systems, and compare their BLEU scores on these sets.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Hyperparameters Our model has two Dirichlet concentration hyperparameters: \u00ce\u00b1 is the shared hyperparameter for the token-level HMM emission and transition distributions.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Table 4 shows translation results for the three approaches.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ex: Mr. Cristiani is the president ...","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We have shown that the maximum entropy framework is able to use global information directly.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For example, hanzi containing the INSECT radical !R tend to denote insects and other crawling animals; examples include tr wal 'frog,' feng1 'wasp,' and !Itt she2 'snake.'","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The English data comes from the WSJ portion of the Penn Treebank and the other languages from the training set of the CoNLL-X multilingual dependency parsing shared task.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For each domain, we created a semantic dictionary by doing two things.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For IE, the system must be able to distinguish between semantically similar noun phrases that play different roles in an event.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In approaching this problem, a variety of different methods are conceivable, including a more or less sophisticated use of machine learning.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Segmental morphology Hebrew consists of seven particles m(\u201cfrom\u201d) f(\u201cwhen\u201d\/\u201cwho\u201d\/\u201cthat\u201d) h(\u201cthe\u201d) w(\u201cand\u201d) k(\u201clike\u201d) l(\u201cto\u201d) and b(\u201cin\u201d). which may never appear in isolation and must always attach as prefixes to the following open-class category item we refer to as stem.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"This aspect of the formalism is both linguistically and computationally important.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"By applying an inverse transformation to the output of the parser, arcs with non-standard labels can be lowered to their proper place in the dependency graph, giving rise 1The dependency graph has been modified to make the final period a dependent of the main verb instead of being a dependent of a special root node for the sentence. to non-projective structures.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Only IRSTLM does not support threading.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"A total of 13,976 phrases were grouped.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"In (Bikel et al., 1997) and (Bikel et al., 1999), performance was plotted against training data size to show how performance improves with training data size.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Ideally we\u00e2\u0080\u0099d like to know the thematic role of each extracted noun phrase, but AutoSlog does not generate thematic roles.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Because the Bikel parser has been parameter- ized for Arabic by the LDC, we do not change the default model settings.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"To prevent this we &quot;smooth&quot; the confidence by adding a small value, e, to both W+ and W_, giving at = Plugging the value of at from Equ.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"It differs from previous machine learning-based NERs in that it uses information from the whole document to classify each word, with just one classifier.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"60 ( P e tr o v, 2 0 0 9 ) all B e r k e l e y ( S e p . 0 9 ) B a s e l i n e 7 0 a l l G o l d P O S 70 \u00e2\u0080\u0094 \u00e2\u0080\u0094 \u00e2\u0080\u0094 0 . 8 0 9 0.839 335 0 . 7 9","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Realizing gains in practice can be challenging, however, particularly when the target domain is distant from the background data.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"As we have noted in Section 2, the general semantic class to which a hanzi belongs is often predictable from its semantic radical.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Our new approach, which is called quasi-monotone search, processes the source sentence monotonically, while explicitly taking into account the positions of the German verbgroup.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The baseline system in Table 3 refers to the maximum entropy system that uses only local features.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Morphological analyzers for Hebrew that analyze a surface form in isolation have been proposed by Segal (2000), Yona and Wintner (2005), and recently by the knowledge center for processing Hebrew (Itai et al., 2006).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We also removed the context-type feature type when using the CoBoost approach.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We proposed an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This is a simple and effective alternative to setting weights discriminatively to maximize a metric such as BLEU.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"It is not easy to make a clear definition of \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If evidence indicates that hypotheses C and D are less likely than hypotheses A and B, then probabilities are redistributed to reflect the fact that {A, B} is more likely to contain the answer than {C, D}.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"36 79.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"In our experiments, we used the same set of features as BergKirkpatrick et al. (2010): an indicator feature based In a traditional Markov model, the emission distribution P\u0398(Xi = xi  |Zi = zi) is a set of multinomials.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We call this approach parser switching.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Different sentence structure and rich target language morphology are two reasons for this.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"vierten 12.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Proper names are assumed to be coreferent if they match exactly, or if they closely match based on a few heuristics.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Judges varied in the average score they handed out.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The average individual parser accuracy was reduced by more than 5% when we added this new parser, but the precision of the constituent voting technique was the only result that decreased significantly.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Word frequencies are estimated by a re-estimation procedure that involves apply\u00c2\u00ad ing the segmentation algorithm presented here to a corpus of 20 million words,8 using 8 Our training corpus was drawn from a larger corpus of mixed-genre text consisting mostly of.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The sentences in the corpus were tagged by a transformation-based chunker and an NE tagger.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"(2006) developed a technique for splitting and chunking long sentences.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"3.4 Salience-based text generation.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"One of such approaches uses comparable documents, which are sets of documents whose content are found\/known to be almost the same, such as different newspaper stories about the same event [Shinyama and Sekine 03] or different translations of the same story [Barzilay 01].","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Syntactic decoders, such as cdec (Dyer et al., 2010), build state from null context then store it in the hypergraph node for later extension.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"However there is no global pruning.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"'Malaysia.'","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"So, there is a limitation that IE can only be performed for a predefined task, like \u00e2\u0080\u009ccorporate mergers\u00e2\u0080\u009d or \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"We follow the guidelines developed in the TIGER project (Brants et al. 2002) for syntactic annotation of German newspaper text, using the Annotate3 tool for interactive construction of tree structures.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"If e < b then the key is not found.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Thus we opted not to take the step of creating more precise written annotation guidelines (as (Carlson, Marcu 2001) did for English), which would then allow for measuring inter-annotator agreement.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The dev corpus was taken from the NIST05 evaluation set, augmented with some randomly-selected material reserved from the training set.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Comparison with state-of-the-art taggers For comparison we consider two unsupervised tag- gers: the HMM with log-linear features of Berg- Kirkpatrick et al.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Figure 4 shows the seven general knowledge sources (KSs) that represent features commonly used for coreference resolution.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"To bridge this gap, we consider a practically motivated scenario, in which we want to leverage existing resources from a resource-rich language (like English) when building tools for resource-poor foreign languages.1 We assume that absolutely no labeled training data is available for the foreign language of interest, but that we have access to parallel data with a resource-rich language.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The PCFG was trained from the same sections of the Penn Treebank as the other three parsers.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"86 78.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"During training, we treat as observed the language word types W as well as the token-level corpus w. We utilize Gibbs sampling to approximate our collapsed model posterior: P (T ,t|W , w, \u00ce\u00b1, \u00ce\u00b2) \u00e2\u0088\u009d P (T , t, W , w|\u00ce\u00b1, \u00ce\u00b2) 0.7 0.6 0.5 0.4 0.3 English Danish Dutch Germany Portuguese Spanish Swedish = P (T , t, W , w, \u00cf\u0088, \u00ce\u00b8, \u00cf\u0086, w|\u00ce\u00b1, \u00ce\u00b2)d\u00cf\u0088d\u00ce\u00b8d\u00cf\u0086 Note that given tag assignments T , there is only one setting of token-level tags t which has mass in the above posterior.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The goal of our research was to explore the use of contextual role knowledge for coreference resolution.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"wo rd => na m e 2.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"(c) After they blindfolded the men...","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The above auxiliary quantity satisfies the following recursive DP equation: Qe0 (e; C; j) = Initial Skip Verb Final 1.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A second restriction on the formalisms is that choices during the derivation are independent of the context in the derivation.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"7 www.cis.upenn.edu\/\u00e2\u0088\u00bcpdtb\/ 8 www.eml-research.de\/english\/Research\/NLP\/ Downloads had to buy a new car.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Same for . In the case where the next token is a hyphen, then is also used as a feature: (init- Caps, ) is set to 1.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"One is the accuracy within a set of phrases which share the same keyword; the other is the accuracy of links.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Several papers report the use of part-of-speech information to rank segmentations (Lin, Chiang, and Su 1993; Peng and Chang 1993; Chang and Chen 1993); typically, the probability of a segmentation is multiplied by the probability of the tagging(s) for that segmentation to yield an estimate of the total probability for the analysis.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"But we also need an estimate of the probability for a non-occurring though possible plural form like i\u00c2\u00a5JJ1l.f, nan2gua1-men0 'pumpkins.'","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Recently, this topic has been getting more attention, as is evident from the Paraphrase Workshops in 2003 and 2004, driven by the needs of various NLP applications.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We have developed a coreference resolver called BABAR that uses contextual role knowledge to make coreference decisions.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For the examples given in (1) and (2) this certainly seems possible.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The 8 similarity-to-IN features are based on word frequencies and scores from various models trained on the IN corpus: To avoid numerical problems, each feature was normalized by subtracting its mean and dividing by its standard deviation.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Table 6: Incremental dev set results for the manually annotated grammar (sentences of length \u00e2\u0089\u00a4 70).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Compared with SRILM, IRSTLM adds several features: lower memory consumption, a binary file format with memory mapping, caching to increase speed, and quantization.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The system of Berg-Kirkpatrick et al.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"To explore this tradeoff, we have performed experiments with three different encoding schemes (plus a baseline), which are described schematically in Table 1.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Finally, we note that Jiang\u2019s instance-weighting framework is broader than we have presented above, encompassing among other possibilities the use of unlabelled IN data, which is applicable to SMT settings where source-only IN corpora are available.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Then each arc of D maps either from an element of H to an element of p, or from E-i.e., the empty string-to an element of P. More specifically, each word is represented in the dictionary as a sequence of arcs, starting from the initial state of D and labeled with an element 5 of Hxp, which is terminated with a weighted arc labeled with an element of Ex P. The weight represents the estimated cost (negative log probability) of the word.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The dictionary sizes reported in the literature range from 17,000 to 125,000 entries, and it seems reasonable to assume that the coverage of the base dictionary constitutes a major factor in the performance of the various approaches, possibly more important than the particular set of methods used in the segmentation.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Base NPs are the other significant category of nominal phrases.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"1.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For example, in Northern dialects (such as Beijing), a full tone (1, 2, 3, or 4) is changed to a neutral tone (0) in the final syllable of many words: Jll donglgual 'winter melon' is often pronounced donglguaO.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The points enumerated above are particularly related to ITS, but analogous arguments can easily be given for other applications; see for example Wu and Tseng's (1993) discussion of the role of segmentation in information retrieval.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"While Gan's system incorporates fairly sophisticated models of various linguistic information, it has the drawback that it has only been tested with a very small lexicon (a few hundred words) and on a very small test set (thirty sentences); there is therefore serious concern as to whether the methods that he discusses are scalable.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"The basic word order is VSO, but SVO, VOS, and VO configurations are also possible.2 Nouns and verbs are created by selecting a consonantal root (usually triliteral or quadriliteral), which bears the semantic core, and adding affixes and diacritics.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Ablation Analysis We evaluate the impact of incorporating various linguistic features into our model in Table 3.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This leads to a linear combination of domain-specific probabilities, with weights in [0, 1], normalized to sum to 1.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The tool we use is MMAX8, which has been specifically designed for marking co-reference.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"If there are too many distinct states, the decoder prunes low-scoring partial hypotheses, possibly leading to a search error.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"We have described grammar state splits that significantly improve parsing performance, catalogued parsing errors, and quantified the effect of segmentation errors.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In the appositive case, the contextual predictor was the head of the modifying appositive (president in the Maury Cooper example); in the second case, the contextual predictor was the preposition together with the noun it modifies (plant_in in the Georgia example).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"We can do that . IbmS: Yes, wonderful.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Further, the special hash 0 suffices to flag empty buckets.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This makes memory usage comparable to our PROBING model.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Output of the learning algorithm: a function h:Xxy [0, 1] where h(x, y) is an estimate of the conditional probability p(y1x) of seeing label y given that feature x is present.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"These are shown, with their associated costs, as follows: ABj nc 4.0 AB C\/jj 6.0 CD \/vb 5.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"(6), with W+ > W_.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We measured recall (Rec), precision (Pr), and the F-measure (F) with recall and precision equally weighted.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Thus we are interested not in extraction, but actual generation from representations that may be developed to different degrees of granularity.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"For a description of the application of AdaBoost to various NLP problems see the paper by Abney, Schapire, and Singer in this volume.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"IRSTLM (Federico et al., 2008) is an open-source toolkit for building and querying language models.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"While our method also enforces a singe tag per word constraint, it leverages the transition distribution encoded in an HMM, thereby benefiting from a richer representation of context.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"For example, in the CC-domain, 96 keywords are found which have TF\/ITF scores above a threshold; some of them are shown in Figure 3.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"A Brief Introduction to the Chinese Writing System Most readers will undoubtedly be at least somewhat familiar with the nature of the Chinese writing system, but there are enough common misunderstandings that it is as well to spend a few paragraphs on properties of the Chinese script that will be relevant to topics discussed in this paper.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"lla\/llb and 14a\/14b respectively).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We have provided methods for handling certain classes of unknown words, and models for other classes could be provided, as we have noted.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"2.3 Rhetorical structure.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"A different knowledge source, called CFSemCFSem, compares the semantic expectations of the caseframe that extracts the anaphor with the semantic expectations of the caseframe that extracts the candidate.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"For displaying and querying the annoated text, we make use of the Annis Linguistic Database developed in our group for a large research effort (\u00e2\u0080\u0098Sonderforschungsbereich\u00e2\u0080\u0099) revolving around 9 2.7 Information structure.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The alignment mapping is j ! i = aj from source position j to target position i = aj . The use of this alignment model raises major problems if a source word has to be aligned to several target words, e.g. when translating German compound nouns.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Better Arabic Parsing: Baselines, Evaluations, and Analysis","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"On average, 6 reference translations per automatic translation are available.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"IRST is not threadsafe.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Limitations There are several limitations in the methods.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"It differs from previous machine learning-based NERs in that it uses information from the whole document to classify each word, with just one classifier.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We have already mentioned the closely related work by Matsoukas et al (2009) on discriminative corpus weighting, and Jiang and Zhai (2007) on (nondiscriminative) instance weighting.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Our analysis identifies three key factors driving our performance gain: 1) selecting a model structure which directly encodes tag sparsity, 2) a type-level prior on tag assignments, and 3) a straightforward na\u00c2\u00a8\u00c4\u00b1veBayes approach to incorporate features.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(b) 89 :1 t& tal de cai2neng2 hen3 he DE talent very 'He has great talent' f.b ga ol hig h While the current algorithm correctly handles the (b) sentences, it fails to handle the (a) sentences, since it does not have enough information to know not to group the sequences.ma3lu4 and?]cai2neng2 respectively.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Previous work deals with this problem by correcting inconsistencies between the named entity classes assigned to different occurrences of the same entity (Borthwick, 1999; Mikheev et al., 1998).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"A direct-mapped cache makes BerkeleyLM faster on repeated queries, but their fastest (scrolling) cached version is still slower than uncached PROBING, even on cache-friendly queries.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Model Overview The model starts by generating a tag assignment T for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Finally, the statistical method fails to correctly group hanzi in cases where the individual hanzi comprising the name are listed in the dictionary as being relatively high-frequency single-hanzi words.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Furthermore, by inverting the transducer so that it maps from phonemic transcriptions to hanzi sequences, one can apply the segmenter to other problems, such as speech recognition (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"We have presented a method for unsupervised part- of-speech tagging that considers a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Thus, if one wants to segment words-for any purpose-from Chinese sentences, one faces a more difficult task than one does in English since one cannot use spacing as a guide.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The aforementioned surface form bcl, for example, may also stand for the lexical item \u201conion\u201d, a Noun.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank, leaving only sections 22 and 23 completely untouched during the development of any of the parsers.","label":1},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"These results are achieved by training on the official MUC6 and MUC7 training data, which is much less training data than is used by other machine learning systems that worked on the MUC6 or MUC7 named entity task (Bikel et al., 1997; Bikel et al., 1999; Borth- wick, 1999).","label":1},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"In that work, mutual information was used to decide whether to group adjacent hanzi into two-hanzi words.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Models To assess the marginal utility of each component of the model (see Section 3), we incremen- tally increase its sophistication.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For example, if FCC and Federal Communications Commission are both found in a document, then Federal has A begin set to 1, Communications has A continue set to 1, Commission has A end set to 1, and FCC has A unique set to 1.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Both authors are members of the Center for Language and Speech Processing at Johns Hopkins University.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Hence, the different averages of manual scores for the different language pairs reflect the behaviour of the judges, not the quality of the systems on different language pairs.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Because the Bikel parser has been parameter- ized for Arabic by the LDC, we do not change the default model settings.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For English POS tagging, BergKirkpatrick et al. (2010) found that this direct gradient method performed better (>7% absolute accuracy) than using a feature-enhanced modification of the Expectation-Maximization (EM) algorithm (Dempster et al., 1977).8 Moreover, this route of optimization outperformed a vanilla HMM trained with EM by 12%.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Compared to decoding, this task is cache-unfriendly in that repeated queries happen only as they naturally occur in text.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"We have presented a new method for non-projective dependency parsing, based on a combination of data-driven projective dependency parsing and graph transformation techniques.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":".., Wn ) (obs) P T : Tag assigns (T1 ,.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"A promising direction for future work is to explicitly model a distribution over tags for each word type.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The Verbmobil task is an appointment scheduling task.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We report micro-averaged (whole corpus) and macro-averaged (per sentence) scores along add a constraint on the removal of punctuation, which has a single tag (PUNC) in the ATB.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"However, the data sparsity induced by vocalization makes it difficult to train statistical models on corpora of the size of the ATB, so vocalizing and then parsing may well not help performance.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We report the F1 value of both measures.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The problem is to store these two values for a large and sparse set of n-grams in a way that makes queries efficient.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"4 Evaluation Results.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Depending on the threshold t0, the search algorithm may miss the globally optimal path which typically results in additional translation errors.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In total there are O(K 2) parameters associated with the transition parameters.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"The ATB is disadvantaged by having fewer trees with longer average 5 LDC A-E catalog numbers: LDC2008E61 (ATBp1v4), LDC2008E62 (ATBp2v3), and LDC2008E22 (ATBp3v3.1).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We are also grateful to Markus Dickinson, Ali Farghaly, Nizar Habash, Seth Kulick, David McCloskey, Claude Reichard, Ryan Roth, and Reut Tsarfaty for constructive discussions.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"SRILM\u2019s compact variant has an incredibly expensive destructor, dwarfing the time it takes to perform translation, and so we also modified Moses to avoiding the destructor by calling exit instead of returning normally.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"If they are found in a list, then a feature for that list will be set to 1.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Proper names that match are resolved with each other.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Based on revision 4041, we modified Moses to print process statistics before terminating.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Their theoretical finding is simply stated: classification error rate decreases toward the noise rate exponentially in the number of independent, accurate classifiers.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"= p(fj je) max \u00c3\u0086;e00 j02Cnfjg np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j 0 )o: The DP equation is evaluated recursively for each hypothesis (e0; e; C; j).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Gazdar (1985) argues that sharing of stacks can be used to give analyses for coordination.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Thus, for example, one successor process will be have M to be in the existential state qa with the indices encoding xi , , xn, in the first 2n i tapes.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Unfortunately, Yarowsky's method is not well understood from a theoretical viewpoint: we would like to formalize the notion of redundancy in unlabeled data, and set up the learning task as optimization of some appropriate objective function.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"We have provided methods for handling certain classes of unknown words, and models for other classes could be provided, as we have noted.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Thus we opted not to take the step of creating more precise written annotation guidelines (as (Carlson, Marcu 2001) did for English), which would then allow for measuring inter-annotator agreement.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"2 61.7 64.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"However, MADA is language-specific and relies on manually constructed dictionaries.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank, leaving only sections 22 and 23 completely untouched during the development of any of the parsers.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The original tag set for the CoNLL-X Dutch data set consists of compounded tags that are used to tag multi-word units (MWUs) resulting in a tag set of over 300 tags.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Still, from a theoretical point of view, projective parsing of non-projective structures has the drawback that it rules out perfect accuracy even as an asymptotic goal.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Finally, we provide a realistic eval uation in which segmentation is performed both in a pipeline and jointly with parsing (\u00c2\u00a76).","label":1},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"AdaBoost is given access to a weak learning algorithm, which accepts as input the training examples, along with a distribution over the instances.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"MUC7 has also seen hybrids of statistical NERs and hand-coded systems (Mikheev et al., 1998; Borthwick, 1999), notably Mikheev' s system, which achieved the best performance of 93.39% on the official NE test data.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The second model (+PRIOR) utilizes the independent prior over type-level tag assignments P (T |\u00cf\u0088).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Not surprisingly some semantic classes are better for names than others: in our corpora, many names are picked from the GRASS class but very few from the SICKNESS class.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"To prevent this we &quot;smooth&quot; the confidence by adding a small value, e, to both W+ and W_, giving at = Plugging the value of at from Equ.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"However, this result is consistent with the results of ex\u00c2\u00ad periments discussed in Wu and Fung (1994).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The procedure using the tagged sentences to discover paraphrases takes about one hour on a 2GHz Pentium 4 PC with 1GB of memory.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We call this approach parser switching.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Since the transducers are built from human-readable descriptions using a lexical toolkit (Sproat 1995), the system is easily maintained and extended.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"3.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Mutual information was shown to be useful in the segmentation task given that one does not have a dictionary.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"(If fewer than n rules have Precision greater than pin, we 3Note that taking tlie top n most frequent rules already makes the method robut to low count events, hence we do not use smoothing, allowing low-count high-precision features to be chosen on later iterations. keep only those rules which exceed the precision threshold.) pm,n was fixed at 0.95 in all experiments in this paper.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"In this paper, we offer broad insight into the underperformance of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The phrases have to be the expressions of length less than 5 chunks, appear between two NEs.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"For robustness reasons, the parser may output a set of dependency trees instead of a single tree. most dependent of the next input token, dependency type features are limited to tokens on the stack.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We are interested in combining the substructures of the input parses to produce a better parse.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"A receives a votes, and B receives b votes.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"All four of the techniques studied result in parsing systems that perform better than any previously reported.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For terrorism, BABAR generated 5,078 resolutions: 2,386 from lexical seeding and 2,692 from syntactic seeding.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We again assume a training set of n examples {x1 . xri} where the first m examples have labels {y1 ... yin}, and the last (n \u2014 m) examples are unlabeled.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Given an anaphor and candidate, BABAR checks (1) whether the semantic classes of the anaphor intersect with the semantic expectations of the caseframe that extracts the candidate, and (2) whether the semantic classes of the candidate intersect with the semantic ex pectations of the caseframe that extracts the anaphor.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Finally, we add \u00e2\u0080\u009cDT\u00e2\u0080\u009d to the tags for definite nouns and adjectives (Kulick et al., 2006).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"!!:\\ :yu2 e:_nc [::!!:zen3 l!f :moO t:_adv il!:shuot ,:_vb i i i 1 \u00e2\u0080\u00a2 10.03 13...","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"\u00e2\u0080\u009cThe gun\u00e2\u0080\u009d will be extracted by the caseframe \u00e2\u0080\u009cfired <patient>\u00e2\u0080\u009d.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Table 2 compares the performance of our system on the setup of Cohen and Smith (2007) to the best results reported by them for the same tasks.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For instance, the common \"suffixes,\" -nia (e.g.,.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The PROBING model was designed to improve upon SRILM by using linear probing hash tables (though not arranged in a trie), allocating memory all at once (eliminating the need for full pointers), and being easy to compile.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Denote by g3(x) = Et crithl(x) , j E {1,2} the unthresholded strong-hypothesis (i.e., f3 (x) = sign(gi (x))).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The transitive closure of the dictionary in (a) is composed with Id(input) (b) to form the WFST (c).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Note that in our model, conditioned on T , there is precisely one t which has nonzero probability for the token component, since for each word, exactly one \u00ce\u00b8t has support.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"D o m ai n # of ph ras es t o t a l p h r a s e s ac cu ra cy C C 7 o r m o r e 1 0 5 8 7 . 6 % 6 o r l e s s 1 0 6 6 7 . 0 % P C 7 o r m o r e 3 5 9 9 9 . 2 % 6 o r l e s s 2 5 5 6 5 . 1 % Table 1.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In this work, we take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"All language model queries issued by machine translation decoders follow a left-to-right pattern, starting with either the begin of sentence token or null context for mid-sentence fragments.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In general, different modalities (\u00e2\u0080\u009cplanned to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cagreed to buy\u00e2\u0080\u009d, \u00e2\u0080\u009cbought\u00e2\u0080\u009d) were considered to express the same relationship within an extraction setting.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"This result suggests the benefit of using the automatic discovery method.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The distinctions in the ATB are linguistically justified, but complicate parsing.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The second experiment is limited to data from PDT.5 The training part of the treebank was projectivized under different encoding schemes and used to train memory-based dependency parsers, which were run on the test part of the treebank, consisting of 7,507 sentences and 125,713 tokens.6 The inverse transformation was applied to the output of the parsers and the result compared to the gold standard test set.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We then gather all phrases with the same keyword.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"(3)In sentence (1), McCann can be a person or an orga nization.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Now, for this application one might be tempted to simply bypass the segmentation problem and pronounce the text character-by-character.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Section 2 describes our baseline techniques for SMT adaptation, and section 3 describes the instance-weighting approach.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"\u2014 I would also like to point out to commissioner Liikanen that it is not easy to take a matter to a national court.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"raphy: A ren2 'person' is a fairly uncontroversial case of a monographemic word, and rplil zhong1guo2 (middle country) 'China' a fairly uncontroversial case of a di\u00c2\u00ad graphernic word.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Similarly, there is no compelling evidence that either of the syllables of f.ifflll binllang2 'betelnut' represents a morpheme, since neither can occur in any context without the other: more likely fjfflll binllang2 is a disyllabic morpheme.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"1 61.2 43.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Hyperparameters Our model has two Dirichlet concentration hyperparameters: \u00ce\u00b1 is the shared hyperparameter for the token-level HMM emission and transition distributions.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Both BBN and NYU have tagged their own data to supplement the official training data.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Maximizing (7) is thus much faster than a typical MERT run. where co(s, t) are the counts from OUT, as in (6).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Each out-of-domain phrase pair is characterized by a set of simple features intended to reflect how useful it will be.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"All annotations are done with specific tools and in XML; each layer has its own DTD.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Table 2 shows BABAR\u00e2\u0080\u0099s performance.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The highestorder N-gram array omits backoff and the index, since these are not applicable.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This implies, therefore, that a major factor in the performance of a Chinese segmenter is the quality of the base dictionary, and this is probably a more important factor-from the point of view of performance alone-than the particular computational methods used.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"A procedural definition to restrict1In the approach described in (Berger et al., 1996), a mor phological analysis is carried out and word morphemes rather than full-form words are used during the search.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, if {N P1, N P2, N P3} are all coreferent, then each NP must be linked to one of the other two NPs.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Parent Head Modif er Dir # gold F1 Label # gold F1 NP NP TAG R 946 0.54 ADJP 1216 59.45 S S S R 708 0.57 SBAR 2918 69.81 NP NP ADJ P R 803 0.64 FRAG 254 72.87 NP NP N P R 2907 0.66 VP 5507 78.83 NP NP SBA R R 1035 0.67 S 6579 78.91 NP NP P P R 2713 0.67 PP 7516 80.93 VP TAG P P R 3230 0.80 NP 34025 84.95 NP NP TAG L 805 0.85 ADVP 1093 90.64 VP TAG SBA R R 772 0.86 WHN P 787 96.00 S VP N P L 961 0.87 (a) Major phrasal categories (b) Major POS categories (c) Ten lowest scoring (Collins, 2003)-style dependencies occurring more than 700 times Table 8: Per category performance of the Berkeley parser on sentence lengths \u00e2\u0089\u00a4 70 (dev set, gold segmentation).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"This approach is compared to another reordering scheme presented in (Berger et al., 1996).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"It uses a maximum entropy framework and classifies each word given its features.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"These tag distributions are used to initialize the label distributions over the English vertices in the graph.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"token-level HMM to reflect lexicon sparsity.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"For the \u00e2\u0080\u0098core\u00e2\u0080\u0099 portion of PCC, we found that on average, 35% of the coherence relations in our RST annotations are explicitly signalled by a lexical connective.6 When adding the fact that connectives are often ambiguous, one has to conclude that prospects for an automatic analysis of rhetorical structure using shallow methods (i.e., relying largely on connectives) are not bright \u00e2\u0080\u0094 but see Sections 3.2 and 3.3 below.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Interpolation search has a more expensive pivot but performs less pivoting and reads, so it is slow on small data and faster on large data.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The goal of machine translation is the translation of a text given in some source language into a target language.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"In practice, this sparsity constraint is difficult to incorporate in a traditional POS induction system (Me\u00c2\u00b4rialdo, 1994; Johnson, 2007; Gao and Johnson, 2008; Grac\u00c2\u00b8a et al., 2009; Berg-Kirkpatrick et al., 2010).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"This model admits a simple Gibbs sampling algorithm where the number of latent variables is proportional to the number of word types, rather than the size of a corpus as for a standard HMM sampler (Johnson, 2007).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The ATB has a much higher fraction of nuclei per tree, and a higher type-level error rate.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"If the expression is longer or complicated (like \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d and \u00e2\u0080\u009cA\u00e2\u0080\u0099s purchase of B\u00e2\u0080\u009d), it is called \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d, i.e. a set of phrases which express the same thing or event.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The domains are the linguistic spans that are to receive an IS-partitioning, and the units are the (smaller) spans that can play a role as a constituent of such a partitioning.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"Machine Translation In this case my colleague can not visit on I n d i e s e m F a l l ka nn m e i n K o l l e g e a m the v i e r t e n M a i n i c h t b e s u c h e n S i e you fourth of May Figure 1: Reordering for the German verbgroup.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Maximizing (7) is thus much faster than a typical MERT run. where co(s, t) are the counts from OUT, as in (6).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"An important aspect of the DempsterShafer model is that it operates on sets of hypotheses.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Mai.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"As noted in Section 4.4, disk cache state is controlled by reading the entire binary file before each test begins.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"For queries, we uniformly sampled 10 million hits and 10 million misses.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In Table 5 we present results from small test cor\u00c2\u00ad pora for the productive affixes handled by the current version of the system; as with names, the segmentation of morphologically derived words is generally either right or wrong.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The translation search is carried out with the category markers and the city names are resubstituted into the target sentence as a postprocessing step.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For example, a story can mention \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d, \u00e2\u0080\u009cthe White House\u00e2\u0080\u009d, or \u00e2\u0080\u009cthe weather\u00e2\u0080\u009d without any prior referent in the story.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The algorithm, called CoBoost, has the advantage of being more general than the decision-list learning alInput: (xi , yi), , (xim, ) ; x, E 2x, yi = +1 Initialize Di (i) = 1\/m.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.6 Co-reference.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Location list is processed into a list of unigrams and bigrams (e.g., New York).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"First we present the non-parametric version of parser switching, similarity switching: The intuition for this technique is that we can measure a similarity between parses by counting the constituents they have in common.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Sima\u2019an et al. (2001) presented parsing results for a DOP tree-gram model using a small data set (500 sentences) and semiautomatic morphological disambiguation.","label":0},{"source_text":"There is no global pruning.","target_text":"Pr(eI 1) is the language model of the target language, whereas Pr(fJ 1 jeI1) is the transla tion model.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We smooth Prf(p \u2014* (s, p)) for rare and OOV segments (s E l, l E L, s unseen) using a \u201cper-tag\u201d probability distribution over rare segments which we estimate using relative frequency estimates for once-occurring segments.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The details are given in (Tillmann, 2000).","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The accuracy results for segmentation, tagging and parsing using our different models and our standard data split are summarized in Table 1.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Syntactic universals are a well studied concept in linguistics (Carnie, 2002; Newmeyer, 2005), and were recently used in similar form by Naseem et al. (2010) for multilingual grammar induction.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Each word is terminated by an arc that represents the transduction between f and the part of speech of that word, weighted with an estimated cost for that word.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Further, it needs extra pointers in the trie, increasing model size by 40%.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"One obvious application is information extraction.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"As a result, Habash et al.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"In the named entity task, X1 might be the instance space for the spelling features, X2 might be the instance space for the contextual features.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(7), such as the likelihood function used in maximum-entropy problems and other generalized additive models (Lafferty 99).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"First, we parsed the training corpus, collected all the noun phrases, and looked up each head noun in WordNet (Miller, 1990).","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"With the additional assumption that (s, t) can be restricted to the support of co(s, t), this is equivalent to a \u201cflat\u201d alternative to (6) in which each non-zero co(s, t) is set to one.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"In each figure the upper graph shows the isolated constituent precision and the bottom graph shows the corresponding number of hypothesized constituents.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":", December, then the feature MonthName is set to 1.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The path set of a tree set is the union of the path sets of trees in that tree set.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For example, in Figure 3, we can see that the phrases in the \u00e2\u0080\u009cbuy\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire\u00e2\u0080\u009d and \u00e2\u0080\u009cpurchase\u00e2\u0080\u009d sets are mostly paraphrases.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"First, we learn weights on individual phrase pairs rather than sentences.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Pairwise comparison is done using the sign test.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"We show how a datadriven deterministic dependency parser, in itself restricted to projective structures, can be combined with graph transformation techniques to produce non-projective structures.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Overview of the method 2.2 Step by Step Algorithm.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"com \u00c2\u00a7Cambridge, UK Email: nc201@eng.cam.ac.uk \u00c2\u00a9 1996 Association for Computational Linguistics (a) B ) ( , : & ; ? ' H o w d o y o u s a y o c t o p u s i n J a p a n e s e ? ' (b) P l a u s i b l e S e g m e n t a t i o n I B X I I 1 : & I 0 0 r i 4 w e n 2 z h a n g l y u 2 z e n 3 m e 0 s h u o l ' J a p a n e s e ' ' o c t o p u s ' ' h o w ' ' s a y ' (c) Figure 1 I m p l a u s i b l e S e g m e n t a t i o n [\u00c2\u00a7] lxI 1:&I ri4 wen2 zhangl yu2zen3 me0 shuol 'Japan' 'essay' 'fish' 'how' 'say' A Chinese sentence in (a) illustrating the lack of word boundaries.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The task can be considered to be one component of the MUC (MUC-6, 1995) named entity task (the other task is that of segmentation, i.e., pulling possible people, places and locations from text before sending them to the classifier).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"If no candidate satisfies this condition (which is often the case), then the anaphor is left unresolved.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"They are: 5We are grateful to an anonymous reviewer for pointing this out.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Corpus Step 1 NE pair instances Step 2 Step 1.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The algorithm can be viewed as heuristically optimizing an objective function suggested by (Blum and Mitchell 98); empirically it is shown to be quite successful in optimizing this criterion.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Lexical-knowledge-based approaches that include statistical information generally presume that one starts with all possible segmentations of a sentence, and picks the best segmentation from the set of possible segmentations using a probabilistic or cost\u00c2\u00ad based scoring mechanism.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"However, the next step is clearly different.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","label":1},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Sometimes extracted phrases by themselves are not meaningful to consider without context, but we set the following criteria.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The evidence may not say anything about whether A is more likely than B, only that C and D are not likely.","label":0},{"source_text":"There is no global pruning.","target_text":"In this section, we brie y review our translation approach.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Interestingly, Chang et al. report 80.67% recall and 91.87% precision on an 11,000 word corpus: seemingly, our system finds as many names as their system, but with four times as many false hits.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"In the first stage, we run a single step of label propagation, which transfers the label distributions from the English vertices to the connected foreign language vertices (say, Vf\ufffd) at the periphery of the graph.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Bresnan, Kaplan, Peters, and Zaenen (1982) argue that these structures are needed to describe crossed-serial dependencies in Dutch subordinate clauses.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"They are: 5We are grateful to an anonymous reviewer for pointing this out.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Location list is processed into a list of unigrams and bigrams (e.g., New York).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"So, who won the competition?","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"raphy: A ren2 'person' is a fairly uncontroversial case of a monographemic word, and rplil zhong1guo2 (middle country) 'China' a fairly uncontroversial case of a di\u00c2\u00ad graphernic word.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Table 3 contains the results for evaluating our systems on the test set (section 22).","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Assuming unseen objects within each class are equiprobable, their probabilities are given by the Good-Turing theorem as: cis E( n'J.ls) Po oc N * E(N8ls) (2) where p815 is the probability of one unseen hanzi in class cls, E(n'J.15 ) is the expected number of hanzi in cls seen once, N is the total number of hanzi, and E(N(\/ 5 ) is the expected number of unseen hanzi in class cls.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"There have been other kinds of efforts to discover paraphrase automatically from corpora.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The Potsdam Commentary Corpus","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"123 examples fell into the noise category.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Each feature group can be made up of many binary features.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This differs from other implementations (Stolcke, 2002; Pauls and Klein, 2011) that use hash tables as nodes in a trie, as explained in the next section.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For example, take ..., says Maury Cooper, a vice president at S.&P.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Sequence of Initial Caps (SOIC): In the sentence Even News Broadcasting Corp., noted for its accurate reporting, made the erroneous announcement., a NER may mistake Even News Broadcasting Corp. as an organization name.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Mutual information was shown to be useful in the segmentation task given that one does not have a dictionary.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"How do additional ambiguities caused by devocalization affect statistical learning?","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"0 55.3 34.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Otherwise, it is set to 0.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"computing the recall of the other's judgments relative to this standard.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In order to train this error-correction model, he divided his training corpus into 5 portions of 20% each.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"In Table 7 we give results for several evaluation metrics.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In Section 2, we brie y review our approach to statistical machine translation.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In Section 3, we introduce our novel concept to word reordering and a DP-based search, which is especially suitable for the translation direction from German to English.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This means that the rules in our grammar are of two kinds: (a) syntactic rules relating nonterminals to a sequence of non-terminals and\/or PoS tags, and (b) lexical rules relating PoS tags to lattice arcs (lexemes).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Frequency is calculated by counting the number of distinct previous tokens that each token has (e.g., if Electric Corp. is seen 3 times, and Manufacturing Corp. is seen 5 times during training, and Corp. is not seen with any other preceding tokens, then the \u00e2\u0080\u009cfrequency\u00e2\u0080\u009d of Corp. is 2).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"40 75.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(1991}, Gu and Mao (1994), and Nie, Jin, and Hannan (1994).","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"There are two differences between this method and the DL-CoTrain algorithm: spelling and contextual features, alternating between labeling and learning with the two types of features.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"We show translation results for three approaches: the monotone search (MonS), where no word reordering is allowed (Tillmann, 1997), the quasimonotone search (QmS) as presented in this paper and the IBM style (IbmS) search as described in Section 3.2.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The first unsupervised algorithm we describe is based on the decision list method from (Yarowsky 95).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"f1; ; Jg denotes a coverage set including all positions from the starting position 1 to position J and j 2 fJ \u00f4\u0080\u0080\u0080L; ; Jg.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Our analysis and comparison focuses primarily on the one-to-one accuracy since it is a stricter metric than many-to-one accuracy, but also report many-to-one for completeness.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For example, as Gan (1994) has noted, one can construct examples where the segmen\u00c2\u00ad tation is locally ambiguous but can be determined on the basis of sentential or even discourse context.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"First of all, most previous articles report perfor\u00c2\u00ad mance in terms of a single percent-correct score, or else in terms of the paired measures of precision and recall.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For irt the Good-Turing estimate just discussed gives us an estimate of p(unseen(f,) I f,)-the probability of observing a previously unseen instance of a construction in ft given that we know that we have a construction in f,.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Each ht is a function that predicts a label (+1 or \u20141) on examples containing a particular feature xt, while abstaining on other examples: The prediction of the strong hypothesis can then be written as We now briefly describe how to choose ht and at at each iteration.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"M spawns as many processes as there are ways of breaking up ri , .. \u2022 , zt, and rules with A on their left-hand-side.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The supervised POS tagging accuracies (on this tagset) are shown in the last row of Table 2.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"3.2 Reordering with IBM Style.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Third, all remaining anaphora are evaluated by 11 different knowledge sources: the four contextual role knowledge sources just described and seven general knowledge sources.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"An extension of the TAG system was introduced by Joshi et al. (1975) and later redefined by Joshi (1987) in which the adjunction operation is defined on sets of elementary trees rather than single trees.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Within the RST \u00e2\u0080\u009cuser community\u00e2\u0080\u009d there has also been discussion whether two levels of discourse structure should not be systematically distinguished (intentional versus informational).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Before explaining our method in detail, we present a brief overview in this subsection.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Affix Pron Base category N found N missed (recall) N correct (precision) t,-,7 The second issue is that rare family names can be responsible for overgeneration, especially if these names are otherwise common as single-hanzi words.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"When using the segmentation pruning (using HSPELL) for unseen tokens, performance improves for all tasks as well.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Others depend upon various lexical heuris\u00c2\u00ad tics: for example Chen and Liu (1992) attempt to balance the length of words in a three-word window, favoring segmentations that give approximately equal length for each word.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"All notions of word, with the exception of the orthographic word, are as relevant in Chinese as they are in English, and just as is the case in other languages, a word in Chinese may correspond to one or more symbols in the orthog 1 For a related approach to the problem of word-segrnention in Japanese, see Nagata (1994), inter alia..","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The cost of storing these averages, in bits, is Because there are comparatively few unigrams, we elected to store them byte-aligned and unquantized, making every query faster.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Arguably this consists of about three phonological words.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"An important subproblem of language model storage is therefore sparse mapping: storing values for sparse keys using little memory then retrieving values given keys using little time.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Fort= 1,...,T:","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"One side of the decision making process is when we choose to believe a constituent should be in the parse, even though only one parser suggests it.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Finally, the statistical method fails to correctly group hanzi in cases where the individual hanzi comprising the name are listed in the dictionary as being relatively high-frequency single-hanzi words.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"If the context wnf will never extend to the right (i.e. wnf v is not present in the model for all words v) then no subsequent query will match the full context.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Second, rather than relying on a division of the corpus into manually-assigned portions, we use features intended to capture the usefulness of each phrase pair.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"pre-processing.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"If so, the CF Network reports that the anaphor and candidate may be coreferent.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Each connected path (l1 ... lk) E L corresponds to one morphological segmentation possibility of W. The Parser Given a sequence of input tokens W = w1 ... wn and a morphological analyzer, we look for the most probable parse tree \u03c0 s.t.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The difference in performance between pronouns and definite noun phrases surprised us.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"For example, one parser could be more accurate at predicting noun phrases than the other parsers.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"This approach leads to a search procedure with complexity O(E3 J4).","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We train linear mixture models for conditional phrase pair probabilities over IN and OUT so as to maximize the likelihood of an empirical joint phrase-pair distribution extracted from a development set.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"In fact, training on the official training data is not suitable as the articles in this data set are entirely about aviation disasters, and the test data is about air vehicle launching.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"There any many techniques for improving language model speed and reducing memory consumption.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This WFST is then summed with the WFST implementing the dictionary and morphological rules, and the transitive closure of the resulting transducer is computed; see Pereira, Riley, and Sproat (1994) for an explanation of the notion of summing WFSTs.12 Conceptual Improvements over Chang et al.'s Model.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Table 2 shows BABAR\u00e2\u0080\u0099s performance.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"3 68.9 50.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"He used an additional maximum entropy classifier that tries to correct mistakes by using reference resolution.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The proof is given in (Tillmann, 2000).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This would result in better rest cost estimation and better pruning.10 In general, tighter, but well factored, integration between the decoder and language model should produce a significant speed improvement.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"2.2.1 The Caseframe Representation Information extraction (IE) systems use extraction patterns to identify noun phrases that play a specific role in 1 Our implementation only resolves NPs that occur in the same document, but in retrospect, one could probably resolve instances of the same existential NP in different documents too.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"(student+plural) 'students,' which is derived by the affixation of the plural affix f, menD to the nounxue2shengl.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"These methods demonstrated the benefits of incorporating linguistic features using a log-linear parameterization, but requires elaborate machinery for training.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"If the key distribution\u2019s range is also known (i.e. vocabulary identifiers range from 0 to the number of words), then interpolation search can use this information instead of reading A[0] and A[|A |\u2212 1] to estimate pivots; this optimization alone led to a 24% speed improvement.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Although the ultimate goal of the Verbmobil project is the translation of spoken language, the input used for the translation experiments reported on in this paper is the (more or less) correct orthographic transcription of the spoken sentences.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"com t 600 Mountain Avenue, 2c278, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In the natural disasters domain, agents are often forces of nature, such as hurricanes or wildfires.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"For the LM, adaptive weights are set as follows: where \u03b1 is a weight vector containing an element \u03b1i for each domain (just IN and OUT in our case), pi are the corresponding domain-specific models, and \u02dcp(w, h) is an empirical distribution from a targetlanguage training corpus\u2014we used the IN dev set for this.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In any event, to date, we have not compared different methods for deriving the set of initial frequency estimates.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"On several languages, we report performance exceeding that of more complex state-of-the art systems.1","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"TIS systems in general need to do more than simply compute the.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"2 for the accuracy of the different methods.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"The parser switching oracle is the upper bound on the accuracy that can be achieved on this set in the parser switching framework.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The probability of the classes assigned to the words in a sentence in a document is defined as follows: where is determined by the maximum entropy classifier.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For 2 < n < N, we use a hash table mapping from the n-gram to the probability and backoff3.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"In (b) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the kidnapping victims, but in (c) \u00e2\u0080\u009cthey\u00e2\u0080\u009d refers to the armed men.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"68 95.","label":0},{"source_text":"There is no global pruning.","target_text":"Using the concept of inverted alignments, we explicitly take care of the coverage constraint by introducing a coverage set C of source sentence positions that have been already processed.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This step ensures that the most frequent terms for each domain are labeled (in case some of them are not in WordNet) and labeled with the sense most appropriate for the domain.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"t(i).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In the CC-domain, there are 32 sets of phrases which contain more than 2 phrases.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Not only is concatenation of these strings possible, but head wrapping can be used to split a string and wrap it around another string.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"It was also proposed to allow annotators to skip sentences that they are unable to judge.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"BABAR achieved recall in the 4250% range for both domains, with 76% precision overall for terrorism and 87% precision for natural disasters.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Thus at each iteration the method induces at most n x k rules, where k is the number of possible labels (k = 3 in the experiments in this paper). step 3.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Word Head Of Complement POS 1 '01 inna \u00e2\u0080\u009cIndeed, truly\u00e2\u0080\u009d VP Noun VBP 2 '01 anna \u00e2\u0080\u009cThat\u00e2\u0080\u009d SBAR Noun IN 3 01 in \u00e2\u0080\u009cIf\u00e2\u0080\u009d SBAR Verb IN 4 01 an \u00e2\u0080\u009cto\u00e2\u0080\u009d SBAR Verb IN Table 1: Diacritized particles and pseudo-verbs that, after orthographic normalization, have the equivalent surface form 0 an.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For example, if is found in the list of person first names, the feature PersonFirstName is set to 1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Foreign names are usually transliterated using hanzi whose sequential pronunciation mimics the source language pronunciation of the name.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Handling Unknown tokens When handling unknown tokens in a language such as Hebrew various important aspects have to be borne in mind.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Because we are working with only three parsers, the only situation in which context will help us is when it can indicate we should choose to believe a single parser that disagrees with the majority hypothesis instead of the majority hypothesis itself.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"As a result, CFG's can not provide the structural descriptions in which there are nested dependencies between symbols labelling a path.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"However, reads in the TRIE data structure are more expensive due to bit-level packing, so we found that it is faster to use interpolation search the entire time.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In this paper, our goal has been to use the notion of LCFRS's to classify grammatical systems on the basis of their strong generative capacity.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Formally, let el (62) be the number of classification errors of the first (second) learner on the training data, and let Eco be the number of unlabeled examples on which the two classifiers disagree.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Such ambiguities cause discrepancies between token boundaries (indexed as white spaces) and constituent boundaries (imposed by syntactic categories) with respect to a surface form.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The second concerns the methods used (if any) to ex\u00c2\u00ad tend the lexicon beyond the static list of entries provided by the machine-readable dictionary upon which it is based.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Otherwise for the predecessor search hypothesis, we would have chosen a position that would not have been among the first n uncovered positions.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Note also that the costs currently used in the system are actually string costs, rather than word costs.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The sentences in the corpus were tagged by a transformation-based chunker and an NE tagger.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Feature-based HMM Model (Berg- Kirkpatrick et al., 2010): The KM model uses a variety of orthographic features and employs the EM or LBFGS optimization algorithm; Posterior regulariation model (Grac\u00c2\u00b8a et al., 2009): The G10 model uses the posterior regular- ization approach to ensure tag sparsity constraint.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Sc opi ng filters candidate if outside the anaphor\u00e2\u0080\u0099s scope.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"If two nouns have mutually exclusive semantic classes, then they cannot be coreferent.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"0 Figure 5 An example of affixation: the plural affix.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"By establishing significantly higher parsing baselines, we have shown that Arabic parsing performance is not as poor as previously thought, but remains much lower than English.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"There are still some open issues to be resolved with the format, but it represents a first step.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Using these states, we define partial hypothesis extensions, which are of the following type: (S0;C n fjg; j0) !","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"13.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The second row represents the performance of the median hyperparameter setting.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Each knowledge source then assigns a probability estimate to each candidate, which represents its belief that the candidate is the antecedent for the anaphor.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Furthermore, we know one of the original parses will be the hypothesized parse, so the direct method of determining which one is best is to compute the probability of each of the candidate parses using the probabilistic model we developed in Section 2.1.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"3.1 Word ReOrdering with Verbgroup.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"A search restriction especially useful for the translation direction from German to English is presented.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The second algorithm builds on a boosting algorithm called AdaBoost (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The advantage is that we can recombine search hypotheses by dynamic programming.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(e.g., N.Y. would contribute this feature, IBM would not). nonalpha=x Appears if the spelling contains any characters other than upper or lower case letters.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Even with less training data, MENERGI outperforms Borthwick' s MENE + reference resolution (Borthwick, 1999).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We have described two data structures for language modeling that achieve substantial reductions in time and memory cost.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Several systems propose statistical methods for handling unknown words (Chang et al. 1992; Lin, Chiang, and Su 1993; Peng and Chang 1993).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"While we have minimized forward-looking state in Section 4.1, machine translation systems could also benefit by minimizing backward-looking state.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For corporate suffixes, a list of tokens cslist that occur frequently as the last token of an organization name is collected from the training data.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Backoff-smoothed models estimate this probability based on the observed entry with longest matching history wnf , returning where the probability p(wn|wn\u22121 f ) and backoff penalties b(wn\u22121 i ) are given by an already-estimated model.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Manual annotation results in human in- terpretable grammars that can inform future tree- bank annotation decisions.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"annotation guidelines that tell annotators what to do in case of doubt.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Further, it needs extra pointers in the trie, increasing model size by 40%.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Two subjects are each given a calendar and they are asked to schedule a meeting.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"We implement two data structures: PROBING, designed for speed, and TRIE, optimized for memory.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The key to the methods we describe is redundancy in the unlabeled data.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"For seven out of eight languages a threshold of 0.2 gave the best results for our final model, which indicates that for languages without any validation set, r = 0.2 can be used.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The 1st block contains the simple baselines from section 2.1.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Altun et al. (2005) proposed a technique that uses graph based similarity between labeled and unlabeled parts of structured data in a discriminative framework for semi-supervised learning.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Otherwise, it is set to 0.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Before we turn to the evaluation, however, we need to introduce the data-driven dependency parser used in the latter experiments.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"For the LM, adaptive weights are set as follows: where \u03b1 is a weight vector containing an element \u03b1i for each domain (just IN and OUT in our case), pi are the corresponding domain-specific models, and \u02dcp(w, h) is an empirical distribution from a targetlanguage training corpus\u2014we used the IN dev set for this.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Our implementation permits jumping to any n-gram of any length with a single lookup; this appears to be unique among language model implementations.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The likelihood of the observed data under the model is where P(yi, xi) is defined as in (9).","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Note that in our construction arcs can never cross token boundaries.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Further, it needs extra pointers in the trie, increasing model size by 40%.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We first define &quot;pseudo-labels&quot;,-yt, as follows: = Yi t sign(g 0\\ 2\u2014 kx2,m < i < n Thus the first m labels are simply copied from the labeled examples, while the remaining (n \u2014 m) examples are taken as the current output of the second classifier.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"A position is presented by the word at that position.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"We have estimated the performance of IdentiFinder ' 99 at 200K words of training data from the graphs.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Second, BABAR performs reliable case resolution to identify anaphora that can be easily resolved using the lexical and syntactic heuristics described in Section 2.1.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Notice that the CC-domain is a special case.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"A morphological analyzer M : W\u2014* L is a function mapping sentences in Hebrew (W E W) to their corresponding lattices (M(W) = L E L).","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"(f1; ;mg n fl1; l2g ; l) 4 (f1; ;m \u00f4\u0080\u0080\u0080 1g n fl1; l2; l3g ; l0) !","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The contextual role knowledge that BABAR uses for coreference resolution is derived from this caseframe data.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Out of those 15 links, 4 are errors, namely \u00e2\u0080\u009cbuy - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cacquire - pay\u00e2\u0080\u009d, \u00e2\u0080\u009cpurchase - stake\u00e2\u0080\u009d \u00e2\u0080\u009cacquisition - stake\u00e2\u0080\u009d.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"CHARACTERIZING STRUCTURAL DESCRIPTIONS PRODUCED BY VARIOUS GRAMMATICAL FORMALISMS*","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"We show how a datadriven deterministic dependency parser, in itself restricted to projective structures, can be combined with graph transformation techniques to produce non-projective structures.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":" We evaluate the system's performance by comparing its segmentation 'Tudgments\" with the judgments of a pool of human segmenters, and the system is shown to perform quite well.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":1},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The treebank has two versions, v1.0 and v2.0, containing 5001 and 6501 sentences respectively.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"This WFST represents the segmentation of the text into the words AB and CD, word boundaries being marked by arcs mapping between f and part-of-speech labels.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Of course, since the number of attested (phonemic) Mandarin syllables (roughly 1400, including tonal distinctions) is far smaller than the number of morphemes, it follows that a given syllable could in principle be written with any of several different hanzi, depending upon which morpheme is intended: the syllable zhongl could be lfl 'middle,''clock,''end,' or ,'loyal.'","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This is a simple and effective alternative to setting weights discriminatively to maximize a metric such as BLEU.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The SynRole KS computes the relative frequency with which the candidates\u00e2\u0080\u0099 syntactic role (subject, direct object, PP object) appeared in resolutions in the training set.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Microsoft\u2019s approach uses dependency trees, others use hierarchical phrase models.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"See Table 2 for the tag set size of other languages.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2009).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The local features used are similar to those used in BBN' s IdentiFinder (Bikel et al., 1999) or MENE (Borthwick, 1999).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Table 1: Syntactic Seeding Heuristics BABAR\u00e2\u0080\u0099s reliable case resolution heuristics produced a substantial set of anaphor\/antecedent resolutions that will be the training data used to learn contextual role knowledge.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Methods for expanding the dictionary include, of course, morphological rules, rules for segmenting personal names, as well as numeral sequences, expressions for dates, and so forth (Chen and Liu 1992; Wang, Li, and Chang 1992; Chang and Chen 1993; Nie, Jin, and Hannan 1994).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Specifically, we assume each word type W consists of feature-value pairs (f, v).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"We tagged each noun with the top-level semantic classes assigned to it in Word- Net.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"One hybridization strategy is to let the parsers vote on constituents' membership in the hypothesized set.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"However, we needed to restrict ourselves to these languages in order to be able to evaluate the performance of our approach.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The first row represents the average accuracy of the three parsers we combine.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For example, from the sentence \u00e2\u0080\u009cMr.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Although this is not a precise criterion, most cases we evaluated were relatively clear-cut.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In a more recent study than Chang et al., Wang, Li, and Chang (1992) propose a surname-driven, non-stochastic, rule-based system for identifying personal names.17 Wang, Li, and Chang also compare their performance with Chang et al.'s system.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"MADA uses an ensemble of SVMs to first re-rank the output of a deterministic morphological analyzer.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The model is often further restricted so that each source word is assigned to exactly one target word (Brown et al., 1993; Ney et al., 2000).","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"Table 5 shows the overall parsing accuracy attained with the three different encoding schemes, compared to the baseline (no special arc labels) and to training directly on non-projective dependency graphs.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Hence, trees shown in Figure 8 can not be generated by any MCTAG (but can be generated by an IG) because the number of pairs of dependent paths grows with n. Since the derivation tees of TAG's, MCTAG's, and HG's are local sets, the choice of the structure used at each point in a derivation in these systems does not depend on the context at that point within the derivation.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The AdaBoost algorithm was developed for supervised learning.","label":1},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The focus of this work is on building POS taggers for foreign languages, assuming that we have an English POS tagger and some parallel text between the two languages.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"8 1 8.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The domains are the linguistic spans that are to receive an IS-partitioning, and the units are the (smaller) spans that can play a role as a constituent of such a partitioning.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"However, when the linear LM is combined with a linear TM (lm+lin tm) or MAP TM (lm+map TM), the results are much better than a log-linear combination for the EMEA setting, and on a par for NIST.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"In this section, we extend state to optimize left-to-right queries.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The English data comes from the WSJ portion of the Penn Treebank and the other languages from the training set of the CoNLL-X multilingual dependency parsing shared task.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Finally, our U (unparsed) measure is used to report the number of sentences to which our system could not propose a joint analysis.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The key to the methods we describe is redundancy in the unlabeled data.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Backoff-smoothed models estimate this probability based on the observed entry with longest matching history wnf , returning where the probability p(wn|wn\u22121 f ) and backoff penalties b(wn\u22121 i ) are given by an already-estimated model.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"4 Traditional Arabic linguistic theory treats both of these types as subcategories of noun \u00ef\u00bf\u00bd '.i . Figure 1: The Stanford parser (Klein and Manning, 2002) is unable to recover the verbal reading of the unvocalized surface form 0 an (Table 1).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The first method uses a similar algorithm to that of (Yarowsky 95), with modifications motivated by (Blum and Mitchell 98).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"If 0(A) gives the number of occurrences of each terminal in the structure named by A, then, given the constraints imposed on the formalism, for each rule A --. fp(Ai, , An) we have the equality where c\u201e is some constant.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Applications such as machine translation use language model probability as a feature to assist in choosing between hypotheses.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"It is also true of the adaptation of the Collins parser for Czech (Collins et al., 1999) and the finite-state dependency parser for Turkish by Oflazer (2003).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"If enough parsers suggest that a particular constituent belongs in the parse, we include it.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Table 4 shows how much the Bayes switching technique uses each of the parsers on the test set.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Finally, Section 5 explains how BABAR relates to previous work, and Section 6 summarizes our conclusions.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The basic word order is VSO, but SVO, VOS, and VO configurations are also possible.2 Nouns and verbs are created by selecting a consonantal root (usually triliteral or quadriliteral), which bears the semantic core, and adding affixes and diacritics.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Precision is the portion of hypothesized constituents that are correct and recall is the portion of the Treebank constituents that are hypothesized.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"From this we see that a finer-grained model for parser combination, at least for the features we have examined, will not give us any additional power.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Unlabeled examples in the named-entity classification problem can reduce the need for supervision to a handful of seed rules.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Then, token- level HMM emission parameters are drawn conditioned on these assignments such that each word is only allowed probability mass on a single assigned tag.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"This number must be less than or equal to n \u00f4\u0080\u0080\u0080 1.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"A Hebrew surface token may have several readings, each of which corresponding to a sequence of segments and their corresponding PoS tags.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In various dialects of Mandarin certain phonetic rules apply at the word.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The first four affixes are so-called resultative affixes: they denote some prop\u00c2\u00ad erty of the resultant state of a verb, as in E7 wang4bu4-liao3 (forget-not-attain) 'cannot forget.'","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"We present two algorithms.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"-1 means that an NP should be ruled out as a possible antecedent, and 0 means that the knowledge source remains neutral (i.e., it has no reason to believe that they cannot be coreferent).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"As we will see from Table 3, not much improvement is derived from this feature.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The final score is obtained from: max e;e0 j2fJ\u00f4\u0080\u0080\u0080L;;Jg p($je; e0) Qe0 (e; I; f1; ; Jg; j); where p($je; e0) denotes the trigram language model, which predicts the sentence boundary $ at the end of the target sentence.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Next, we describe four contextual role knowledge sources that are created from the training examples and the caseframes.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We utilized two kinds of datasets in our experiments: (i) monolingual treebanks9 and (ii) large amounts of parallel text with English on one side.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"For the English RST-annotated corpus that is made available via LDC, his corresponding result is 62%.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"A moment's reflection will reveal that things are not quite that simple.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"(2003), which gives 96.8% accuracy on the test set.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Comparison with state-of-the-art taggers For comparison we consider two unsupervised tag- gers: the HMM with log-linear features of Berg- Kirkpatrick et al.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We train linear mixture models for conditional phrase pair probabilities over IN and OUT so as to maximize the likelihood of an empirical joint phrase-pair distribution extracted from a development set.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Instead, we want to apply an inverse transformation to recover the underlying (nonprojective) dependency graph.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"We incorporate instance weighting into a mixture-model framework, and find that it yields consistent improvements over a wide range of baselines.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"We would like to relax somewhat the constraint on the path complexity of formalisms in LCFRS.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Whether a language even has orthographic words is largely dependent on the writing system used to represent the language (rather than the language itself); the notion \"orthographic word\" is not universal.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Morphological segmentation decisions in our model are delegated to a lexeme-based PCFG and we show that using a simple treebank grammar, a data-driven lexicon, and a linguistically motivated unknown-tokens handling our model outperforms (Tsarfaty, 2006) and (Cohen and Smith, 2007) on the joint task and achieves state-of-the-art results on a par with current respective standalone models.2","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"(2010), we adopt a simpler na\u00c2\u00a8\u00c4\u00b1ve Bayes strategy, where all features are emitted independently.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The model starts by generating a tag assignment for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Similarly, the tokens and are tested against each list, and if found, a corresponding feature will be set to 1.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"83 77.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This has led previous workers to adopt ad hoc linear weighting schemes (Finch and Sumita, 2008; Foster and Kuhn, 2007; L\u00a8u et al., 2007).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In particular, it may not be possible to learn functions fi (x f2(x2,t) for i = m + 1...n: either because there is some noise in the data, or because it is just not realistic to expect to learn perfect classifiers given the features used for representation.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We model po(s|t) using a MAP criterion over weighted phrase-pair counts: and from the similarity to (5), assuming y = 0, we see that w\u03bb(s, t) can be interpreted as approximating pf(s, t)\/po(s, t).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Not all the layers have been produced for all the texts yet.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"For our demonstration system, we typically use the pruning threshold t0 = 5:0 to speed up the search by a factor 5 while allowing for a small degradation in translation accuracy.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In order to observe the similarity between these constrained systems, it is crucial to abstract away from the details of the structures and operations used by the system.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Features and context were initially introduced into the models, but they refused to offer any gains in performance.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"We plan to explore more powerful techniques for exploiting the diversity of parsing methods.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Their default variant implements a forward trie, in which words are looked up in their natural left-to-right order.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Formally, the DempsterShafer theory defines a probability density function m(S), where S is a set of hypotheses.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Our monolingual similarity function (for connecting pairs of foreign trigram types) is the same as the one used by Subramanya et al. (2010).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"In this section, we extend state to optimize left-to-right queries.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"We generate these caseframes automatically by running AutoSlog over the training corpus exhaustively so that it literally generates a pattern to extract every noun phrase in the corpus.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Ltd., then organization will be more probable.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As with personal names, we also derive an estimate from text of the probability of finding a transliterated name of any kind (PTN).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In addition to the tapes required to store the indices, M requires one work tape for splitting the substrings.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In addition, there are several approaches to non-projective dependency parsing that are still to be evaluated in the large (Covington, 1990; Kahane et al., 1998; Duchier and Debusmann, 2001; Holan et al., 2001; Hellwig, 2003).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We will evaluate various specific aspects of the segmentation, as well as the overall segmentation per\u00c2\u00ad formance.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Thus, rather than give a single evaluative score, we prefer to compare the performance of our method with the judgments of several human subjects.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Overall, the difference between our most basic model (1TW) and our full model (+FEATS) is 21.2% and 13.1% for the best and median settings respectively.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Maamouri et al.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The advantage is that we can recombine search hypotheses by dynamic programming.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In our situation, the competing hypotheses are the possible antecedents for an anaphor.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"In focus is in particular the correlation with rhetorical structure, i.e., the question whether specific rhetorical relations \u00e2\u0080\u0094 or groups of relations in particular configurations \u00e2\u0080\u0094 are signalled by speakers with prosodic means.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"But in most cases they can be used interchangably.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The reference medicine for Silapo is EPREX\/ERYPO, which contains epoetin alfa.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The tool we use is MMAX8, which has been specifically designed for marking co-reference.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The computing time is given in terms of CPU time per sentence (on a 450MHz PentiumIIIPC).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"3.1 General Knowledge Sources.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"On the other hand, in a translation system one probably wants to treat this string as a single dictionary word since it has a conventional and somewhat unpredictable translation into English.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"This is similar to stacking the different feature instantiations into long (sparse) vectors and computing the cosine similarity between them.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We can show that languages generated by LCFRS's are semilinear as long as the composition operation does not remove any terminal symbols from its arguments.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"This method, one instance of which we term the \"greedy algorithm\" in our evaluation of our own system in Section 5, involves starting at the beginning (or end) of the sentence, finding the longest word starting (ending) at that point, and then repeating the process starting at the next (previous) hanzi until the end (begin\u00c2\u00ad ning) of the sentence is reached.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"(1998) did make use of information from the whole document.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"It falls short of the \u201cProjection\u201d baseline for German, but is statistically indistinguishable in terms of accuracy.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The family name set is restricted: there are a few hundred single-hanzi family names, and about ten double-hanzi ones.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Most languages that use Roman, Greek, Cyrillic, Armenian, or Semitic scripts, and many that use Indian-derived scripts, mark orthographic word boundaries; however, languages written in a Chinese-derived writ\u00c2\u00ad ing system, including Chinese and Japanese, as well as Indian-derived writing systems of languages like Thai, do not delimit orthographic words.1 Put another way, written Chinese simply lacks orthographic words.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"These words are in turn highly ambiguous, breaking the assumption underlying most parsers that the yield of a tree for a given sentence is known in advance.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"\u2022 Similarly, when the na\u00efve Bayes classifier is configured such that the constituents require estimated probabilities strictly larger than 0.5 to be accepted, there is not enough probability mass remaining on crossing brackets for them to be included in the hypothesis.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"We represent all morphological analyses of a given utterance using a lattice structure.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Among these 32 sets, we found the following pairs of sets which have two or more links.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Such a classification can be seen as a not-always-correct summary of global features.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"1 55.8 38.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"We group the features used into feature groups.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"For the disasters domain, 8245 texts were used for training and the 40 test documents contained 447 anaphoric links.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Apart from MERT difficulties, a conceptual problem with log-linear combination is that it multiplies feature probabilities, essentially forcing different features to agree on high-scoring candidates.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Surprisingly, the non-parametric switching technique also exhibited robust behaviour in this situation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The method being described-henceforth ST..","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"RandLM\u2019s stupid backoff variant stores counts instead of probabilities and backoffs.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"This group consists of (3 total number of possible zones) features.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The model described here thus demonstrates great potential for use in widespread applications.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Linear probing hash tables must have more buckets than entries, or else an empty bucket will never be found.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For example Chen and Liu (1992) report precision and recall rates of over 99%, but this counts only the words that occur in the test corpus that also occur in their dictionary.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"One class comprises words derived by productive morphologi\u00c2\u00ad cal processes, such as plural noun formation using the suffix ir, menD.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Unsupervised Models for Named Entity Classification Collins","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Consider first the examples in (2).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"2 62.6 45.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Each of the constituents must have received at least 1 votes from the k parsers, so a > I1 and 2 \u2014 2k\u00b1-1 b > ri-5-111.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This has the effect of randomly permuting vocabulary identifiers, meeting the requirements of interpolation search when vocabulary identifiers are used as keys.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Our implementation permits jumping to any n-gram of any length with a single lookup; this appears to be unique among language model implementations.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The knowledge base then can be tested for its relation-inference capabilities on the basis of full-blown co-reference information.","label":0},{"source_text":"Here we present two algorithms.","target_text":"This paper discusses the use of unlabeled examples for the problem of named entity classification.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"In fact, training on the official training data is not suitable as the articles in this data set are entirely about aviation disasters, and the test data is about air vehicle launching.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"2.3 Rhetorical structure.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"All commentaries have been tagged with part-of-speech information using Brants\u00e2\u0080\u0099 TnT1 tagger and the Stuttgart\/Tu\u00c2\u00a8bingen Tag Set automatic analysis was responsible for this decision.)","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"In fact, we found that enabling IRSTLM\u2019s cache made it slightly slower, so results in Table 1 use IRSTLM without caching.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The result of this is shown in Figure 7.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"And then there are decisions that systems typically hard-wire, because the linguistic motivation for making them is not well understood yet.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"We substantially outperform all of them on query speed and offer lower memory consumption than lossless alternatives.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"Special thanks to Jan Haji\u02c7c and Matthias Trautner Kromann for assistance with the Czech and Danish data, respectively, and to Jan Haji\u02c7c, Tom\u00b4a\u02c7s Holan, Dan Zeman and three anonymous reviewers for valuable comments on a preliminary version of the paper.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Because these two words have identical complements, syntax rules are typically unhelpful for distinguishing between them.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While it is essential to be fluent in the target language, it is not strictly necessary to know the source language, if a reference translation was given.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"If they are found in a list, then a feature for that list will be set to 1.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"While we used the standard metrics of the community, the we way presented translations and prompted for assessment differed from other evaluation campaigns.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The rationale for treating these semantic labels differently is that they are specific and reliable (as opposed to the WordNet classes, which are more coarse and more noisy due to polysemy).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This model is easily incorporated into the segmenter by building a WFST restrict\u00c2\u00ad ing the names to the four licit types, with costs on the arcs for any particular name summing to an estimate of the cost of that name.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"2.4 Underspecified rhetorical structure.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The PROBING data structure is a rather straightforward application of these hash tables to store Ngram language models.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Note that because we extracted only high-confidence alignments, many foreign vertices will not be connected to any English vertices.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"37.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"0 55.3 34.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"SRILM inefficiently stores 64-bit pointers.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"This is not to say that a set of standards by which a particular segmentation would count as correct and another incorrect could not be devised; indeed, such standards have been proposed and include the published PRCNSC (1994) and ROCLING (1993), as well as the unpublished Linguistic Data Consortium standards (ca.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"It should be clear from the onset that the particle b (\u201cin\u201d) in \u2018bcl\u2019 may then attach higher than the bare noun cl (\u201cshadow\u201d).","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Under certain conditions the constituent voting and na\u00efve Bayes constituent combination techniques are guaranteed to produce sets of constituents with no crossing brackets.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Other approaches encode sparsity as a soft constraint.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"The form mnh itself can be read as at least three different verbs (\u201ccounted\u201d, \u201cappointed\u201d, \u201cwas appointed\u201d), a noun (\u201ca portion\u201d), and a possessed noun (\u201cher kind\u201d).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Figure 5 shows how this model is implemented as part of the dictionary WFST.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"When this feature type was included, CoBoost chose this default feature at an early iteration, thereby giving non-abstaining pseudo-labels for all examples, with eventual convergence to the two classifiers agreeing by assigning the same label to almost all examples.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The type-level posterior term can be computed according to, P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2) \u00e2\u0088\u009d Note that each round of sampling Ti variables takes time proportional to the size of the corpus, as with the standard token-level HMM.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"An initial step of any text\u00c2\u00ad analysis task is the tokenization of the input into words.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"DempsterShafer handles this by re-normalizing all the belief values with respect to only the non-null sets (this is the purpose of the denominator in Equation 1).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Second, we show that although the Penn Arabic Treebank is similar to other tree- banks in gross statistical terms, annotation consistency remains problematic.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Thus an explicit assumption about the redundancy of the features \u2014 that either the spelling or context alone should be sufficient to build a classifier \u2014 has been built into the algorithm.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"State is implemented in their scrolling variant, which is a trie annotated with forward and backward pointers.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"The global feature groups are: InitCaps of Other Occurrences (ICOC): There are 2 features in this group, checking for whether the first occurrence of the same word in an unambiguous position (non first-words in the TXT or TEXT zones) in the same document is initCaps or not-initCaps.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"4.3 Translation Experiments.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Pr(eI 1) is the language model of the target language, whereas Pr(fJ 1 jeI1) is the transla tion model.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The second weakness is purely conceptual, and probably does not affect the per\u00c2\u00ad formance of the model.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The first stage identifies a keyword in each phrase and joins phrases with the same keyword into sets.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Reiche\u00e2\u0080\u0099s colleagues will make sure that the concept is waterproof.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Two issues distinguish the various proposals.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Thus, rather than give a single evaluative score, we prefer to compare the performance of our method with the judgments of several human subjects.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"After tagging a large corpus with an automatic NE tagger, the method tries to find sets of paraphrases automatically without being given a seed phrase or any kind of cue.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Moreover, they are used as substantives much 2 Unlike machine translation, constituency parsing is not significantly affected by variable word order.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In the last example, the less restrictive IbmS word reordering leads to a better translation, although the QmS translation is still acceptable.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"On the one hand, the type-level error rate is not calibrated for the number of n-grams in the sample.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Gan's solution depends upon a fairly sophisticated language model that attempts to find valid syntactic, semantic, and lexical relations between objects of various linguistic types (hanzi, words, phrases).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"As is standard, we use a fixed constant K for the number of tagging states.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Among machine learning-based NERs, Identi- Finder has proven to be the best on the official MUC6 and MUC7 test data.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The following recursive equation is evaluated: Qe0 (e; S; C; j) = (2) = p(fj je) max \u00c3\u0086;e00 np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) max (S0;j0) (S0 ;Cnfjg;j0)!(S;C;j) j02Cnfjg Qe00 (e0; S0;C n fjg; j0)o: The search ends in the hypotheses (I; f1; ; Jg; j).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"So, it is too costly to make IE technology \u00e2\u0080\u009copen- domain\u00e2\u0080\u009d or \u00e2\u0080\u009con-demand\u00e2\u0080\u009d like IR or QA.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"This offers the well-known advantages for inter- changability, but it raises the question of how to query the corpus across levels of annotation.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"For example, out of 905 phrases in the CC- domain, 211 phrases contain keywords found in step 2.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"(f1; ;mg n fl1; l2; l3g ;m) German to English the monotonicity constraint is violated mainly with respect to the German verbgroup.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Dynamic programming efficiently scores many hypotheses by exploiting the fact that an N-gram language model conditions on at most N \u2212 1 preceding words.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The only supervision is in the form of 7 seed rules (namely, that New York, California and U.S. are locations; that any name containing Mr is a person; that any name containing Incorporated is an organization; and that I.B.M. and Microsoft are organizations).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Searching a probing hash table consists of hashing the key, indexing the corresponding bucket, and scanning buckets until a matching key is found or an empty bucket is encountered, in which case the key does not exist in the table.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"But in most cases they can be used interchangably.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Given this similarity function, we define a nearest neighbor graph, where the edge weight for the n most similar vertices is set to the value of the similarity function and to 0 for all other vertices.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The new algorithm, which we call CoBoost, uses labeled and unlabeled data and builds two classifiers in parallel.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"18 77.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Sometimes extracted phrases by themselves are not meaningful to consider without context, but we set the following criteria.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"However, the learning curves in Figure 3 show that the Berkeley parser does not exceed our manual grammar by as wide a margin as has been shown for other languages (Petrov, 2009).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Previous Work.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The normalization factor plays an important role in the AdaBoost algorithm.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Global features are extracted from other occurrences of the same token in the whole document.","label":1},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Limitations of (Blum and Mitchell 98): While the assumptions of (Blum and Mitchell 98) are useful in developing both theoretical results and an intuition for the problem, the assumptions are quite limited.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In Section 6 we dis\u00c2\u00ad cuss other issues relating to how higher-order language models could be incorporated into the model.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The method reported in this paper makes use solely of unigram probabilities, and is therefore a zeroeth-order model: the cost of a particular segmentation is estimated as the sum of the costs of the individual words in the segmentation.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The PCC is not the result of a funded project.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Third, all remaining anaphora are evaluated by 11 different knowledge sources: the four contextual role knowledge sources just described and seven general knowledge sources.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This group of features attempts to capture such information.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"The hyperparameters \u00ce\u00b1 and \u00ce\u00b2 represent the concentration parameters of the token- and type-level components of the model respectively.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"All improvements over the baseline are statistically significant beyond the 0.01 level (McNemar\u2019s test).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The confidence level is then used as the belief value for the knowledge source.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For the English RST-annotated corpus that is made available via LDC, his corresponding result is 62%.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Recently, statistical NERs have achieved results that are comparable to hand-coded systems.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Table 8b shows that verbal nouns are the hardest pre-terminal categories to identify.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"For displaying and querying the annoated text, we make use of the Annis Linguistic Database developed in our group for a large research effort (\u00e2\u0080\u0098Sonderforschungsbereich\u00e2\u0080\u0099) revolving around 9 2.7 Information structure.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"gaolbu4-gaolxing4 (hap-not-happy) 'happy?'","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"All features were conjoined with the state z.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The Levenshtein distance between the automatic translation and each of the reference translations is computed, and the minimum Levenshtein distance is taken.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"However, when the linear LM is combined with a linear TM (lm+lin tm) or MAP TM (lm+map TM), the results are much better than a log-linear combination for the EMEA setting, and on a par for NIST.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Given that weights on all outgoing arcs sum up to one, weights induce a probability distribution on the lattice paths.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"We can model this probability straightforwardly enough with a probabilistic version of the grammar just given, which would assign probabilities to the individual rules.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Out-of-Vocabulary: We derived a lexicon list from WordNet 1.6, and words that are not found in this list have a feature out-of-vocabulary set to 1.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"While the generative power of CG's is greater that of CFG's, it appears to be highly constrained.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"The following three sections elaborate these different stages is more detail.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"On each language we investigate the contribution of each component of our model.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Each visited entry wni stores backoff b(wni ).","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"(f1; ;mg n fl1g ; l) 3 (f1; ;mg n fl; l1; l2g ; l0) !","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We present two algorithms.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"When the connective is an adverbial, there is much less clarity as to the range of the spans.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"We will also directly compare with a baseline similar to the Matsoukas et al approach in order to measure the benefit from weighting phrase pairs (or ngrams) rather than full sentences.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"(a) IDictionary D I D:d\/0.000 B:b\/0.000 B:b\/0.000 ( b ) ( c ) ( d ) I B e s t P a t h ( I d ( I ) o D * ) I cps:nd4.!l(l() Figure 2 An abstract example illustrating the segmentation algorithm.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"4.3 Translation Experiments.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"For each experiment we gave an nonparametric and a parametric technique for combining parsers.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Overall, it gives improvements ranging from 1.1% for German to 14.7% for Italian, for an average improvement of 8.3% over the unsupervised feature-HMM model.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Sequence of Initial Caps (SOIC): In the sentence Even News Broadcasting Corp., noted for its accurate reporting, made the erroneous announcement., a NER may mistake Even News Broadcasting Corp. as an organization name.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"We also performed experiments to evaluate the impact of each type of contextual role knowledge separately.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The most popular approach to dealing with seg\u00c2\u00ad mentation ambiguities is the maximum matching method, possibly augmented with further heuristics.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For the experiments, we used four newswire corpora, the Los Angeles Times\/Washington Post, The New York Times, Reuters and the Wall Street Journal, all published in 1995.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(2006).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(5) and ht into Equ.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"For the \u00e2\u0080\u0098core\u00e2\u0080\u0099 portion of PCC, we found that on average, 35% of the coherence relations in our RST annotations are explicitly signalled by a lexical connective.6 When adding the fact that connectives are often ambiguous, one has to conclude that prospects for an automatic analysis of rhetorical structure using shallow methods (i.e., relying largely on connectives) are not bright \u00e2\u0080\u0094 but see Sections 3.2 and 3.3 below.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The paper explains the design decisions taken in the annotations, and describes a number of applications using this corpus with its multi-layer annotation.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"This design leads to a significant reduction in the computational complexity of training and inference.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The baseline system in Table 3 refers to the maximum entropy system that uses only local features.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"All the NE pair instances which co-occur separated by at most 4 chunks are collected along with information about their NE types and the phrase between the NEs (the \u00e2\u0080\u0098context\u00e2\u0080\u0099).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Recent work has made significant progress on unsupervised POS tagging (Me\u00c2\u00b4rialdo, 1994; Smith and Eisner, 2005; Haghighi and Klein, 2006; Johnson,2007; Goldwater and Griffiths, 2007; Gao and John son, 2008; Ravi and Knight, 2009).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We train and test on the CoNLL-X training set.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This is not to say that a set of standards by which a particular segmentation would count as correct and another incorrect could not be devised; indeed, such standards have been proposed and include the published PRCNSC (1994) and ROCLING (1993), as well as the unpublished Linguistic Data Consortium standards (ca.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"97 81.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"6. 3) all G o l d P O S 7 0 0.7 91 0.825 358 0.7 73 0.818 358 0.8 02 0.836 452 80.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The motivation for our more informal approach was the intuition that there are so many open problems in rhetorical analysis (and more so for German than for English; see below) that the main task is qualitative investigation, whereas rigorous quantitative analyses should be performed at a later stage.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Figure 5 shows how this model is implemented as part of the dictionary WFST.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Crucially, the conventional orthographic form of MSA text is unvocalized, a property that results in a deficient graphical representation.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The Levenshtein distance between the automatic translation and each of the reference translations is computed, and the minimum Levenshtein distance is taken.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In contrast, NNP (proper nouns) form a large portion of vocabulary.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"IRSTLM and BerkeleyLM use this state function (and a limit of N \u22121 words), but it is more strict than necessary, so decoders using these packages will miss some recombination opportunities.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"-1 means that an NP should be ruled out as a possible antecedent, and 0 means that the knowledge source remains neutral (i.e., it has no reason to believe that they cannot be coreferent).","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"However, since we want to preserve as much of the original structure as possible, we are interested in finding a transformation that involves a minimal number of lifts.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The structure uses linear probing hash tables and is designed for speed.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"In total, across all domains, we kept 13,976 phrases with keywords.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"76 16.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The first value reports use immediately after loading while the second reports the increase during scoring. dBerkeleyLM is written in Java which requires memory be specified in advance.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Our baseline for all sentence lengths is 5.23% F1 higher than the best previous result.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Better grammars are shown here to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"This logic applies recursively: if wnf+1 similarly does not extend and has zero log backoff, it too should be omitted, terminating with a possibly empty context.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Instead, we focused on phrases and set the frequency threshold to 2, and so were able to utilize a lot of phrases while minimizing noise.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The BLEU score has been shown to correlate well with human judgement, when statistical machine translation systems are compared (Doddington, 2002; Przybocki, 2004; Li, 2005).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This Good\u00c2\u00ad Turing estimate of p(unseen(f,) If,) can then be used in the normal way to define the probability of finding a novel instance of a construction in ir, in a text: p(unseen(f,)) = p(unseen(f,) I f,) p(fn Here p(ir,) is just the probability of any construction in ft as estimated from the frequency of such constructions in the corpus.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Tsarfaty (2006) argues that for Semitic languages determining the correct morphological segmentation is dependent on syntactic context and shows that increasing information sharing between the morphological and the syntactic components leads to improved performance on the joint task.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"When the same token is to be interpreted as a single lexeme fmnh, it may function as a single adjective \u201cfat\u201d.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"where the husband's family name is optionally prepended to the woman's full name; thus ;f:*lf#i xu3lin2-yan2hai3 would represent the name that Ms. Lin Yanhai would take if she married someone named Xu.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Finally, model GTv = 2 includes parent annotation on top of the various state-splits, as is done also in (Tsarfaty and Sima\u2019an, 2007; Cohen and Smith, 2007).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"This is because in many cases, the use of hyphens can be considered to be optional (e.g., third-quarter or third quarter).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"We report micro-averaged (whole corpus) and macro-averaged (per sentence) scores along add a constraint on the removal of punctuation, which has a single tag (PUNC) in the ATB.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The DL-CoTrain algorithm can be motivated as being a greedy method of satisfying the above 2 constraints.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The similar explanation applies to the link to the \u00e2\u0080\u009cstake\u00e2\u0080\u009d set.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"vierten 12.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The government has to make a decision, and do it quickly.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"We can only compare with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"We use two common techniques, hash tables and sorted arrays, describing each before the model that uses the technique.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Similarly, there is no compelling evidence that either of the syllables of f.ifflll binllang2 'betelnut' represents a morpheme, since neither can occur in any context without the other: more likely fjfflll binllang2 is a disyllabic morpheme.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"An edge indicates that the two features must have the same label.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Backoff-smoothed models estimate this probability based on the observed entry with longest matching history wnf , returning where the probability p(wn|wn\u22121 f ) and backoff penalties b(wn\u22121 i ) are given by an already-estimated model.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Given parameter estimates, the label for a test example x is defined as We should note that the model in equation 9 is deficient, in that it assigns greater than zero probability to some feature combinations that are impossible.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"In other words, the set of hidden states F was chosen to be the fine set of treebank tags.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"(2009), who also incorporate a sparsity constraint, but does via altering the model objective using posterior regularization.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Hereafter, each pair of NE categories will be called a domain; e.g. the \u00e2\u0080\u009cCompany \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain, which we will call CC- domain (Step 2).","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"In all of our experiments, the binary file (whether mapped or, in the case of most other packages, interpreted) is loaded into the disk cache in advance so that lazy mapping will never fault to disk.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"This is akin to PoS tags sequences induced by different parses in the setup familiar from English and explored in e.g.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"attaching to terms denoting human beings.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The horizontal axis in this plot represents the most significant dimension, which explains 62% of the variation.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Since our graph is built from a parallel corpus, we can use standard word alignment techniques to align the English sentences De 5Note that many combinations are impossible giving a PMI value of 0; e.g., when the trigram type and the feature instantiation don\u2019t have words in common. and their foreign language translations Df.6 Label propagation in the graph will provide coverage and high recall, and we therefore extract only intersected high-confidence (> 0.9) alignments De\ufffdf.","label":0},{"source_text":"There is no global pruning.","target_text":"The search starts in the hypothesis (I; f;g; 0).","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"\u00e2\u0080\u0094 similar results have been observed across multiple languages.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"These methods demonstrated the benefits of incorporating linguistic features using a log-linear parameterization, but requires elaborate machinery for training.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The addition of vertical markovization enables non-pruned models to outperform all previously reported re12Cohen and Smith (2007) make use of a parameter (\u03b1) which is tuned separately for each of the tasks.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"An important aspect of the DempsterShafer model is that it operates on sets of hypotheses.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Independence of paths at this level reflects context freeness of rewriting and suggests why they can be recognized efficiently.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"10 Chinese speakers may object to this form, since the suffix f, menD (PL) is usually restricted to.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"However, it is also mostly political content (even if not focused on the internal workings of the European Union) and opinion.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Hereafter, each pair of NE categories will be called a domain; e.g. the \u00e2\u0080\u009cCompany \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain, which we will call CC- domain (Step 2).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The algorithm builds two classifiers iteratively: each iteration involves minimization of a continuously differential function which bounds the number of examples on which the two classifiers disagree.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"First, the training data for the parser is projectivized by applying a minimal number of lifting operations (Kahane et al., 1998) and encoding information about these lifts in arc labels.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Instead, we resort to an iterative update based method.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"By applying an inverse transformation to the output of the parser, arcs with non-standard labels can be lowered to their proper place in the dependency graph, giving rise 1The dependency graph has been modified to make the final period a dependent of the main verb instead of being a dependent of a special root node for the sentence. to non-projective structures.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"Uniform Tag Prior (1TW) Our initial lexicon component will be uniform over possible tag assignments as well as word types.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In the case of adverbial reduplication illustrated in (3b) an adjective of the form AB is reduplicated as AABB.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"We apply a beam search concept as in speech recognition.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"(2008) reported agreement between the teams (measured with Evalb) at 93.8% F1, the level of the CTB.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"0 70.9 42.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Arabic is a morphologically rich language with a root-and-pattern system similar to other Semitic languages.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Most computational models for coreference resolution rely on properties of the anaphor and candidate antecedent, such as lexical matching, grammatical and syntactic features, semantic agreement, and positional information.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Systran submitted their commercial rule-based system that was not tuned to the Europarl corpus.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"gaolxing4 'happy' => F.i'JF.i'J Jl!","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We asked participants to each judge 200\u2013300 sentences in terms of fluency and adequacy, the most commonly used manual evaluation metrics.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For IE, the system must be able to distinguish between semantically similar noun phrases that play different roles in an event.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Papers that use this method or minor variants thereof include Liang (1986), Li et al.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"In order to train this error-correction model, he divided his training corpus into 5 portions of 20% each.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"On several languages, we report performance exceeding that of more complex state-of-the art systems.1","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For 2 < n < N, we use a hash table mapping from the n-gram to the probability and backoff3.","label":0},{"source_text":"There is no global pruning.","target_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In particular, the decision to represent arguments in verb- initial clauses as VP internal makes VSO and VOS configurations difficult to distinguish.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"(Riloff and Jones 99) was brought to our attention as we were preparing the final version of this paper.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Note also that the costs currently used in the system are actually string costs, rather than word costs.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Each unlabeled pair (x1,i, x2,i) is represented as an edge between nodes corresponding to x1,i and X2,i in the graph.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The AdaBoost algorithm was developed for supervised learning.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"It is possible one could produce better models by introducing features describing constituents and their contexts because one parser could be much better than the majority of the others in particular situations.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Also there are cases where one of the two NEs belong to a phrase outside of the relation.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"A novel element of our model is the ability to capture type-level tag frequencies.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Participants and other volunteers contributed about 180 hours of labor in the manual evaluation.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Table 5: Individual Performance of KSs for Disasters (e.g., \u00e2\u0080\u009cthe mayor\u00e2\u0080\u009d vs. \u00e2\u0080\u009cthe journalist\u00e2\u0080\u009d).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Figure 4 shows a constituent headed by a process nominal with an embedded adjective phrase.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For MUC7, there are also no published results on systems trained on only the official training data of 200 aviation disaster articles.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"shortest match at each point.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Each ht is a function that predicts a label (+1 or \u20141) on examples containing a particular feature xt, while abstaining on other examples: The prediction of the strong hypothesis can then be written as We now briefly describe how to choose ht and at at each iteration.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"All language model queries issued by machine translation decoders follow a left-to-right pattern, starting with either the begin of sentence token or null context for mid-sentence fragments.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Though we do not directly compare state implementations, performance metrics in Table 1 indicate our overall method is faster.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"3 60.7 50.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"For one thing, it is not clear who is to receive settlements or what should happen in case not enough teachers accept the offer of early retirement.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"logical rules, and personal names; the transitive closure of the resulting machine is then computed.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The translation direction is from German to English.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"This enables us to build a high performance NER without using separate classifiers to take care of global consistency or complex formulation on smoothing and backoff models (Bikel et al., 1997).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Unfortunately, we were unable to correctly run the IRSTLM quantized variant.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The translation direction is from German to English.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A morpheme, on the other hand, usually corresponds to a unique hanzi, though there are a few cases where variant forms are found.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"We propose maximizing , where is the sequence of named- entity tags assigned to the words in the sentence , and is the information that can be extracted from the whole document containing . Our system is built on a maximum entropy classifier.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"To be a member of LCFRS a formalism must satisfy two restrictions.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"For example, out of 905 phrases in the CC- domain, 211 phrases contain keywords found in step 2.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Lexical-knowledge-based approaches that include statistical information generally presume that one starts with all possible segmentations of a sentence, and picks the best segmentation from the set of possible segmentations using a probabilistic or cost\u00c2\u00ad based scoring mechanism.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"When finished, the whole material is written into an XML-structured annotation file.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The baseline system in Table 3 refers to the maximum entropy system that uses only local features.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Our open-source (LGPL) implementation is also available for download as a standalone package with minimal (POSIX and g++) dependencies.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"AdaBoost.MH maintains a distribution over instances and labels; in addition, each weak-hypothesis outputs a confidence vector with one confidence value for each possible label.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"2 62.6 45.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"3.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"We picked two domains, the CC-domain and the \u00e2\u0080\u009cPerson \u00e2\u0080\u0093 Company\u00e2\u0080\u009d domain (PC-domain), for the evaluation, as the entire system output was too large to evaluate.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"If there are too many distinct states, the decoder prunes low-scoring partial hypotheses, possibly leading to a search error.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Removing the leaves from the resulting tree yields a parse for L under G, with the desired probabilities.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Pumping t2 will change only one branch and leave the other branch unaffected.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"4.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"(a) 1 \u00c2\u00a7 . ;m t 7 leO z h e 4 pil m a 3 lu 4 sh an g4 bi ng 4 t h i s CL (assi fier) horse w ay on sic k A SP (ec t) 'This horse got sick on the way' (b) 1\u00c2\u00a7: . til y zhe4 tiao2 ma3lu4 hen3 shao3 this CL road very few 'Very few cars pass by this road' :$ chel jinglguo4 car pass by 2.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"This assumption, however, is not inherent to type-based tagging models.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"(2009) study related but different multilingual grammar and tagger induction tasks, where it is assumed that no labeled data at all is available.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Given names are most commonly two hanzi long, occasionally one hanzi long: there are thus four possible name types, which can be described by a simple set of context-free rewrite rules such as the following: 1.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Again we notice that the isolated constituent precision is larger than 0.5 only in those partitions that contain very few samples.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Since the parameter and token components will remain fixed throughout experiments, we briefly describe each.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Again we notice that the isolated constituent precision is larger than 0.5 only in those partitions that contain very few samples.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Further, Maamouri and Bies (2004) argued that the English guidelines generalize well to other languages.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Finally, we wish to reiterate an important point.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For example, kidnapping victims should be extracted from the subject of the verb \u00e2\u0080\u009ckidnapped\u00e2\u0080\u009d when it occurs in the passive voice (the shorthand representation of this pattern would be \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Eight out of the thirteen errors in the high frequency phrases in the CC-domain are the phrases in \u00e2\u0080\u009cagree\u00e2\u0080\u009d.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"F1 85 Berkeley 80 Stanford.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The weak hypothesis chosen was then restricted to be a predictor in favor of this label.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"In this case, we have no finite-state restrictions for the search space.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"On the one hand, the definition of composition in Steedman (1985), which technically permits composition of functions with unbounded number of arguments, generates tree sets with dependent paths such as those shown in Figure 6.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For sorted lookup, we compare interpolation search, standard C++ binary search, and standard C++ set based on red-black trees.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Both parameters depend on a single hyperparameter \u00ce\u00b1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Similarly, there is no compelling evidence that either of the syllables of f.ifflll binllang2 'betelnut' represents a morpheme, since neither can occur in any context without the other: more likely fjfflll binllang2 is a disyllabic morpheme.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"By definition, each existential NP uniquely specifies an object or concept, so we can infer that all instances of the same existential NP are coreferent (e.g., \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d always refers to the same entity).","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"2 70.7 52.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"na me =>1 ha nzi fa mi ly 2 ha nzi gi ve n 3.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"For this reason, na\u00efve Bayes classifiers are well-matched to this problem.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"As a (crude) approximation, we normalize the extraction patterns with respect to active and passive voice and label those extractions as agents or patients.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The second algorithm extends ideas from boosting algorithms, designed for supervised learning tasks, to the framework suggested by (Blum and Mitchell 98).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"It is sometimes claimed that one of the advantages of dependency grammar over approaches based on constituency is that it allows a more adequate treatment of languages with variable word order, where discontinuous syntactic constructions are more common than in languages like English (Mel\u2019\u02c7cuk, 1988; Covington, 1990).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The semantic caseframe expectations are used in two ways.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"This left 962 examples, of which 85 were noise.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(In this figure eps is c) be implemented, though, such as a maximal-grouping strategy (as suggested by one reviewer of this paper); or a pairwise-grouping strategy, whereby long sequences of unattached hanzi are grouped into two-hanzi words (which may have some prosodic motivation).","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"These estimates are in turn combined linearly with relative-frequency estimates from an in-domain phrase table.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Many human evaluation metrics have been proposed.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"This line of work has been motivated by empirical findings that the standard EM-learned unsupervised HMM does not exhibit sufficient word tag sparsity.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In that work, mutual information was used to decide whether to group adjacent hanzi into two-hanzi words.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This may be the sign of a maturing research environment.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Upon identifying an anaphoric expression (currently restricted to: pronouns, prepositional adverbs, definite noun phrases), the an- notator first marks the antecedent expression (currently restricted to: various kinds of noun phrases, prepositional phrases, verb phrases, sentences) and then establishes the link between the two.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For MUC6, for example, Corporate- Suffix-List is made up of ltd., associates, inc., co, corp, ltd, inc, committee, institute, commission, university, plc, airlines, co., corp. and Person-Prefix- List is made up of succeeding, mr., rep., mrs., secretary, sen., says, minister, dr., chairman, ms. . For a token that is in a consecutive sequence of init then a feature Corporate-Suffix is set to 1.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Proper names that match are resolved with each other.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Our system does not currently make use of titles, but it would be straightforward to do so within the finite-state framework that we propose.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"We will refer to our system as MENERGI (Maximum Entropy Named Entity Recognizer using Global Information).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We computed BLEU scores for each submission with a single reference translation.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Gazdar (1985) considers a number of linguistic analyses which IG's (but not CFG's) can make, for example, the Norwedish example shown in Figure 1.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"We tokenize MWUs and their POS tags; this reduces the tag set size to 12.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"We thus decided to pay specific attention to them and introduce an annotation layer for connectives and their scopes.","label":0},{"source_text":"There is no global pruning.","target_text":"Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Since each composition operation is linear and nonerasing, a bounded sequences of substrings associated with the resulting structure is obtained by combining the substrings in each of its arguments using only the concatenation operation, including each substring exactly once.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"PoS tags impose a unique morphological segmentation on surface tokens and present a unique valid yield for syntactic trees.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"A defining characteristic of MSA is the prevalence of discourse markers to connect and subordinate words and phrases (Ryding, 2005).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The POS distributions over the foreign trigram types are used as features to learn a better unsupervised POS tagger (\u00a75).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Unigrams also have 64-bit overhead for vocabulary lookup.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Given the bilingual graph described in the previous section, we can use label propagation to project the English POS labels to the foreign language.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"\u00e2\u0080\u00a2 Anaphoric links: the annotator is asked to specify whether the anaphor is a repetition, partial repetition, pronoun, epithet (e.g., Andy Warhol \u00e2\u0080\u0093 the PopArt artist), or is-a (e.g., Andy Warhol was often hunted by photographers.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Among machine learning-based NERs, Identi- Finder has proven to be the best on the official MUC6 and MUC7 test data.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"The 14 general-language features embody straightforward cues: frequency, \u201ccentrality\u201d as reflected in model scores, and lack of burstiness.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"besuchen 9.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"They are set to fixed constants.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"This is a unique object for which we are able to define a proper probability model.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We excluded these from the evaluation as they can be easily identified with a list of days\/months.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For each candidate antecedent, BABAR identifies the caseframe that would extract the candidate, pairs it with the anaphor\u00e2\u0080\u0099s caseframe, and consults the CF Network to see if this pair of caseframes has co-occurred in previous resolutions.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We are not claiming that this method is almighty.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Manual annotation results in human in- terpretable grammars that can inform future tree- bank annotation decisions.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"5 64.7 42.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"By establishing significantly higher parsing baselines, we have shown that Arabic parsing performance is not as poor as previously thought, but remains much lower than English.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Particular relations are also consistent with particular hypotheses about the segmentation of a given sentence, and the scores for particular relations can be incremented or decremented depending upon whether the segmentations with which they are consistent are \"popular\" or not.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"In order to facilitate this task, we extend the set of arc labels to encode information about lifting operations.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"If we were working with more than three parsers we could investigate minority constituents, those constituents that are suggested by at least one parser, but which the majority of the parsers do not suggest.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"For.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"2 61.7 64.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Topicalization of NP subjects in SVO configurations causes confusion with VO (pro-drop).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"This left 962 examples, of which 85 were noise.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Figure 1 shows an example.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Table 3 Classes of words found by ST for the test corpus.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Time starts when Moses is launched and therefore includes model loading time.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"rhetorical analysis We are experimenting with a hybrid statistical and knowledge-based system for discourse parsing and summarization (Stede 2003), (Hanneforth et al. 2003), again targeting the genre of commentaries.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A greedy algorithm (or maximum-matching algorithm), GR: proceed through the sentence, taking the longest match with a dictionary entry at each point.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"There are still some open issues to be resolved with the format, but it represents a first step.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"4 Evaluation Results.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"with the number of exactly matching guess trees.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"This is summarized in Equation 5.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Following the setup of Johnson (2007), we use the whole of the Penn Treebank corpus for training and evaluation on English.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Bikel et al.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The final score is obtained from: max e;e0 j2fJ\u00f4\u0080\u0080\u0080L;;Jg p($je; e0) Qe0 (e; I; f1; ; Jg; j); where p($je; e0) denotes the trigram language model, which predicts the sentence boundary $ at the end of the target sentence.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"10 Here we use the Good-Turing estimate (Baayen 1989; Church and Gale 1991), whereby the aggregate probability of previously unseen instances of a construction is estimated as ni\/N, where N is the total number of observed tokens and n1 is the number of types observed only once.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"To establish a soft correspondence between the two languages, we use a second similarity function, which leverages standard unsupervised word alignment statistics (\u00a73.3).3 Since we have no labeled foreign data, our goal is to project syntactic information from the English side to the foreign side.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Word frequencies are estimated by a re-estimation procedure that involves apply\u00c2\u00ad ing the segmentation algorithm presented here to a corpus of 20 million words,8 using 8 Our training corpus was drawn from a larger corpus of mixed-genre text consisting mostly of.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Our original hope in combining these parsers is that their errors are independently distributed.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"In general, m; l; l0 6= fl1; l2; l3g and in line umber 3 and 4, l0 must be chosen not to violate the above reordering restriction.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Overall, language modeling significantly impacts decoder performance.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The first probability is estimated from a name count in a text database, and the rest of the probabilities are estimated from a large list of personal names.n Note that in Chang et al.'s model the p(rule 9) is estimated as the product of the probability of finding G 1 in the first position of a two-hanzi given name and the probability of finding G2 in the second position of a two-hanzi given name, and we use essentially the same estimate here, with some modifications as described later on.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"f1; ; Jg denotes a coverage set including all positions from the starting position 1 to position J and j 2 fJ \u00f4\u0080\u0080\u0080L; ; Jg.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The standard measures for evaluating Penn Treebank parsing performance are precision and recall of the predicted constituents.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We also mark all tags that dominate a word with the feminine ending :: taa mar buuTa (markFeminine).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The learned information was recycled back into the resolver to improve its performance.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"On the other hand, if it is seen as McCann Pte.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In addition, this formulation results in a dramatic reduction in the number of model parameters thereby, enabling unusually rapid training.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Richer tag sets have been suggested for modeling morphologically complex distinctions (Diab, 2007), but we find that linguistically rich tag sets do not help parsing.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"If so, the CF Network reports that the anaphor and candidate may be coreferent.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(Specifically, the limit n starts at 5 and increases by 5 at each iteration.)","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Process statistics are already collected by the kernel (and printing them has no meaningful impact on performance).","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Table 2 shows our complete set of results.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"In the graphs, system scores are indicated by a point, the confidence intervals by shaded areas around the point.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"The paper explains the design decisions taken in the annotations, and describes a number of applications using this corpus with its multi-layer annotation.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Although this property is not structural, it depends on the structural property that sentences can be built from a finite set of clauses of bounded structure as noted by Joshi (1983\/85).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We propose an unsupervised method to discover paraphrases from a large untagged corpus, without requiring any seed phrase or other cue.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"5.2 Setup.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Commentaries argue in favor of a specific point of view toward some political issue, often dicussing yet dismissing other points of view; therefore, they typically offer a more interesting rhetorical structure than, say, narrative text or other portions of newspapers.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Another technique for parse hybridization is to use a na\u00efve Bayes classifier to determine which constituents to include in the parse.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Two of the Mainlanders also cluster close together but, interestingly, not particularly close to the Taiwan speakers; the third Mainlander is much more similar to the Taiwan speakers.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Annotators suggested that long sentences are almost impossible to judge.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"computing the recall of the other's judgments relative to this standard.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"It is a sequence of proper nouns within an NP; its last word Cooper is the head of the NP; and the NP has an appositive modifier (a vice president at S.&P.) whose head is a singular noun (president).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The implication of this ambiguity for a parser is that the yield of syntactic trees no longer consists of spacedelimited tokens, and the expected number of leaves in the syntactic analysis in not known in advance.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We report token- and type-level accuracy in Table 3 and 6 for all languages and system settings.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For \u00c3\u0086 = 0, no new target word is generated, while an additional source sentence position is covered.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"In Section 4, we present the performance measures used and give translation results on the Verbmobil task.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Figure 4 shows a constituent headed by a process nominal with an embedded adjective phrase.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"This is because in many cases, the use of hyphens can be considered to be optional (e.g., third-quarter or third quarter).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We tagged each noun with the top-level semantic classes assigned to it in Word- Net.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For example, in MUC6, there are four zones (TXT, HL, DATELINE, DD).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The translation of one position in the source sentence may be postponed for up to L = 3 source positions, and the translation of up to two source positions may be anticipated for at most R = 10 source positions.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We train and test on the CoNLL-X training set.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Table 3: Training and test conditions for the Verbmobil task (*number of words without punctuation marks).","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"As we shall argue, the semantic class affiliation of a hanzi constitutes useful information in predicting its properties.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"Sie.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"They are: 5We are grateful to an anonymous reviewer for pointing this out.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Mutual information was shown to be useful in the segmentation task given that one does not have a dictionary.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"immediately by a Romanization into the pinyin transliteration scheme; numerals following each pinyin syllable represent tones.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"3 The Coreference Resolution Model.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The restriction can be expressed in terms of the number of uncovered source sentence positions to the left of the rightmost position m in the coverage set.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"However, using the top-level semantic classes of WordNet proved to be problematic because the class distinctions are too coarse.","label":1},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Our representation of contextual roles is based on information extraction patterns that are converted into simple caseframes.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Values in the trie are minimally sized at the bit level, improving memory consumption over trie implementations in SRILM, IRSTLM, and BerkeleyLM.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"This fact annoyed especially his dog...).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Microsoft\u2019s approach uses dependency trees, others use hierarchical phrase models.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(2006) developed a technique for splitting and chunking long sentences.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In Section 5, we try to compare results of MENE, IdentiFinder, and MENERGI.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The other half was replaced by other participants, so we ended up with roughly the same number.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Handling Unknown tokens When handling unknown tokens in a language such as Hebrew various important aspects have to be borne in mind.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We include a constituent in our hypothesized parse if it appears in the output of a majority of the parsers.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Although this is not a precise criterion, most cases we evaluated were relatively clear-cut.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"In fact, we found that enabling IRSTLM\u2019s cache made it slightly slower, so results in Table 1 use IRSTLM without caching.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Queries take the form p(wn|wn\u22121 1 ) where wn1 is an n-gram.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This is less effective in our setting, where IN and OUT are disparate.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"). context=x The context for the entity.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The implementation is In a similar effort, (G\u00c2\u00a8otze 2003) developed a proposal for the theory-neutral annotation of information structure (IS) \u00e2\u0080\u0094 a notoriously difficult area with plenty of conflicting and overlapping terminological conceptions.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"We place certain restrictions on the composition operations of LCFRS's, restrictions that are shared by the composition operations of the constrained grammatical systems that we have considered.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This is especially effective at reducing load time, since raw bytes are read directly to memory\u2014or, as happens with repeatedly used models, are already in the disk cache.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The price shocked her.), or same-kind (e.g., Her health insurance paid for the hospital fees, but the automobile insurance did not cover the repair.).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"It has been shown for English (Wang and Hirschberg 1992; Hirschberg 1993; Sproat 1994, inter alia) that grammatical part of speech provides useful information for these tasks.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The major problem for our seg\u00c2\u00ad menter, as for all segmenters, remains the problem of unknown words (see Fung and Wu [1994]).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"In the small window on the left, search queries can be entered, here one for an NP that has been annotated on the co-reference layer as bridging.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"We evaluate our approach on eight European languages (\u00a76), and show that both our contributions provide consistent and statistically significant improvements.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Our parsing performance measures (SY N) thus report the PARSEVAL extension proposed in Tsarfaty (2006).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The sentence-selection approach is crude in that it imposes a binary distinction between useful and non-useful parts of OUT.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The normalization factor plays an important role in the AdaBoost algorithm.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Preprocessing the raw trees improves parsing performance considerably.9 We first discard all trees dominated by X, which indicates errors and non-linguistic text.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"We call such a constituent an isolated constituent.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"All our results are obtained by using only the official training data provided by the MUC conferences.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"For example, if Barry is not in commonWords and is found in the list of person first names, then the feature PersonFirstName will be set to 1.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Making the reasonable assumption that similar information is relevant for solving these problems in Chinese, it follows that a prerequisite for intonation-boundary assignment and prominence assignment is word segmentation.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Our code is thread-safe, and integrated into the Moses, cdec, and Joshua translation systems.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"2.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"For instance, on Spanish, the absolute gap on median performance is 10%.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For pronouns, however, all of the knowledge sources increased recall, often substantially, and with little if any decrease in precision.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"We again assume a training set of n examples {x1 . xri} where the first m examples have labels {y1 ... yin}, and the last (n \u2014 m) examples are unlabeled.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Interpolation search formalizes the notion that one opens a dictionary near the end to find the word \u201czebra.\u201d Initially, the algorithm knows the array begins at b +\u2014 0 and ends at e +\u2014 |A|\u22121.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"splitIN captures the verb\/preposition idioms that are widespread in Arabic.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"The prediction based on these features is a knearest neighbor classification, using the IB1 algorithm and k = 5, the modified value difference metric (MVDM) and class voting with inverse distance weighting, as implemented in the TiMBL software package (Daelemans et al., 2003).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"6 Our knowledge sources return some sort of probability estimate, although in some cases this estimate is not especially well-principled (e.g., the Recency KS).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"A more rigid mechanism for modeling sparsity is proposed by Ravi and Knight (2009), who minimize the size of tagging grammar as measured by the number of transition types.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We use N(u) to denote the neighborhood of vertex u, and fixed n = 5 in our experiments.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"In future work, we plan to follow-up on this approach and investigate other ways that contextual role knowledge can be used.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The P (W |T , \u00cf\u0088) term in the lexicon component now decomposes as: n P (W |T , \u00cf\u0088) = n P (Wi|Ti, \u00cf\u0088) i=1 n \u00ef\u00a3\u00ab \u00ef\u00a3\u00b6 tions are not modeled by the standard HMM, which = n \u00ef\u00a3\u00ad n P (v|\u00cf\u0088Ti f )\u00ef\u00a3\u00b8 instead can model token-level frequency.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Section 2.2 then describes our representation for contextual roles and four types of contextual role knowledge that are learned from the training examples.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","label":1},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Recall.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Because of this, we retokenized and lowercased submitted output with our own tokenizer, which was also used to prepare the training and test data.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"8 1 8.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"For example, we may have semantic information (e.g. database query operations) associated with the productions in a grammar.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Vocabulary lookup is a sorted array of 64-bit word hashes.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"3.1 Gross Statistics.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"It was motivated by the observation that the (Yarowsky 95) algorithm added a very large number of rules in the first few iterations.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"We first define &quot;pseudo-labels&quot;,-yt, as follows: = Yi t sign(g 0\\ 2\u2014 kx2,m < i < n Thus the first m labels are simply copied from the labeled examples, while the remaining (n \u2014 m) examples are taken as the current output of the second classifier.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Several coreference resolvers have used supervised learning techniques, such as decision trees and rule learners (Aone and Bennett, 1995; McCarthy and Lehnert, 1995; Ng and Cardie, 2002; Soon et al., 2001).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We refer to (T , W ) as the lexicon of a language and \u00cf\u0088 for the parameters for their generation; \u00cf\u0088 depends on a single hyperparameter \u00ce\u00b2.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Note that Zt is a normalization constant that ensures the distribution Dt+i sums to 1; it is a function of the weak hypothesis ht and the weight for that hypothesis at chosen at the tth round.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"na me =>1 ha nzi fa mi ly 2 ha nzi gi ve n 3.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"For each label (Per s on, organization and Location), take the n contextual rules with the highest value of Count' (x) whose unsmoothed3 strength is above some threshold pmin.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"6 Results and Analysis.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"We assumed that such a contrastive assessment would be beneficial for an evaluation that essentially pits different systems against each other.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Label propagation can then be used to transfer the labels to the peripheral foreign vertices (i.e. the ones adjacent to the English vertices) first, and then among all of the foreign vertices (\u00a74).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Graph construction does not require any labeled data, but makes use of two similarity functions.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Note that in our construction arcs can never cross token boundaries.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"= p(fj je) max \u00c3\u0086;e00 j02Cnfjg np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) Qe00 (e0;C n fjg; j 0 )o: The DP equation is evaluated recursively for each hypothesis (e0; e; C; j).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Table 5 provides insight into the behavior of different models in terms of the tagging lexicon they generate.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"We report micro-averaged (whole corpus) and macro-averaged (per sentence) scores along add a constraint on the removal of punctuation, which has a single tag (PUNC) in the ATB.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Such sequences are given additional features of A begin, A continue, or A end, and the acronym is given a feature A unique.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The second weakness is purely conceptual, and probably does not affect the per\u00c2\u00ad formance of the model.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For MUC6, for example, Corporate- Suffix-List is made up of ltd., associates, inc., co, corp, ltd, inc, committee, institute, commission, university, plc, airlines, co., corp. and Person-Prefix- List is made up of succeeding, mr., rep., mrs., secretary, sen., says, minister, dr., chairman, ms. . For a token that is in a consecutive sequence of init then a feature Corporate-Suffix is set to 1.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The approach gains leverage from natural redundancy in the data: for many named-entity instances both the spelling of the name and the context in which it appears are sufficient to determine its type.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The most frequently occurring last words of organization names in cslist are compiled into a list of corporate suffixes, Corporate-Suffix- List.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The feature-based model replaces the emission distribution with a log-linear model, such that: on the word identity x, features checking whether x contains digits or hyphens, whether the first letter of x is upper case, and suffix features up to length 3.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"A better approach would be to distin guish between these cases, possibly by drawing on the vast linguistic work on Arabic connectives (AlBatal, 1990).","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"This withdrawal by the treasury secretary is understandable, though.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"So we decided to use semantic class information only to rule out candidates.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"This paper describes the several performance techniques used and presents benchmarks against alternative implementations.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"However, to classify a token , while Borthwick uses tokens from to (from two tokens before to two tokens after ), we used only the tokens , , and . Even with local features alone, MENERGI outperforms MENE (Borthwick, 1999).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Unsupervised Part-of-Speech Tagging with Bilingual Graph-Based Projections","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, kidnapping victims should be extracted from the subject of the verb \u00e2\u0080\u009ckidnapped\u00e2\u0080\u009d when it occurs in the passive voice (the shorthand representation of this pattern would be \u00e2\u0080\u009c<subject> were kidnapped\u00e2\u0080\u009d).","label":0},{"source_text":"Here we present two algorithms.","target_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Table 2 Similarity matrix for segmentation judgments.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We further report SYNCS, the parsing metric of Cohen and Smith (2007), to facilitate the comparison.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"Table 3 contains the results for evaluating our systems on the test set (section 22).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Several extensions of AdaBoost for multiclass problems have been suggested (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"9 65.5 46.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Other strategies could readily 6 As a reviewer has pointed out, it should be made clear that the function for computing the best path is. an instance of the Viterbi algorithm.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"e0; e are the last two target words, C is a coverage set for the already covered source positions and j is the last position visited.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"(Other classes handled by the current system are discussed in Section 5.)","label":0},{"source_text":"There is no global pruning.","target_text":"In this paper, we have presented a new, e\u00c3\u0086cient DP-based search procedure for statistical machine translation.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"In this example there are four \"input characters,\" A, B, C and D, and these map respectively to four \"pronunciations\" a, b, c and d. Furthermore, there are four \"words\" represented in the dictionary.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"To facilitate the comparison of our results to those reported by (Cohen and Smith, 2007) we use their data set in which 177 empty and \u201cmalformed\u201d7 were removed.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Trying to integrate constituent ordering and choice of referring expressions, (Chiarcos 2003) developed a numerical model of salience propagation that captures various factors of author\u00e2\u0080\u0099s intentions and of information structure for ordering sentences as well as smaller constituents, and picking appropriate referring expressions.10 Chiarcos used the PCC annotations of co-reference and information structure to compute his numerical models for salience projection across the generated texts.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The first probability is estimated from a name count in a text database, and the rest of the probabilities are estimated from a large list of personal names.n Note that in Chang et al.'s model the p(rule 9) is estimated as the product of the probability of finding G 1 in the first position of a two-hanzi given name and the probability of finding G2 in the second position of a two-hanzi given name, and we use essentially the same estimate here, with some modifications as described later on.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Since our objective is to compare distributions of bracketing discrepancies, we do not use heuristics to prune the set of nuclei.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"It is the performance we could achieve if an omniscient observer told us which parser to pick for each of the sentences.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Our motivation for using DempsterShafer is that it provides a well-principled framework for combining evidence from multiple sources with respect to competing hypotheses.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Also surprising is the low test set OOV rate given the possibility of morphological variation in Arabic.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Two processes are spawned requiring B to derive z,.., and C to derive yi , , y,.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"Out-of-domain test data is from the Project Syndicate web site, a compendium of political commentary.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"This actually happens quite frequently (more below), so that the rankings are broad estimates.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Extract NE instance pairs with contexts First, we extract NE pair instances with their context from the corpus.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"These sequence models-based approaches commonly treat token-level tag assignment as the primary latent variable.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"\u00e2\u0080\u0094ested in correlations between prosody and dis course structure.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Given that weights on all outgoing arcs sum up to one, weights induce a probability distribution on the lattice paths.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Proper-Name Identification.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Statistics for all data sets are shown in Table 2.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For MUC7, there are also no published results on systems trained on only the official training data of 200 aviation disaster articles.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Each production can push or pop symbols on the stack as can be seen in the following productions that generate tree of the form shown in Figure 4b.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Thus we opted not to take the step of creating more precise written annotation guidelines (as (Carlson, Marcu 2001) did for English), which would then allow for measuring inter-annotator agreement.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We then discuss how we adapt and generalize a boosting algorithm, AdaBoost, to the problem of named entity classification.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"If is unique, then a feature (Unique, Zone) is set to 1, where Zone is the document zone where appears.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Each extraction pattern represents a linguistic expression and a syntactic position indicating where a role filler can be found.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Since our destructor is an efficient call to munmap, bypassing the destructor favors only other packages.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Finally, we note that simple weighting gives nearly a 2% F1 improvement, whereas Goldberg and Tsarfaty (2008) found that unweighted lattices were more effective for Hebrew.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"The 2nd block contains the IR system, which was tuned by selecting text in multiples of the size of the EMEA training corpus, according to dev set performance.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Two measures that can be used to compare judgments are: 1.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"First, in section 4, we evaluate the graph transformation techniques in themselves, with data from the Prague Dependency Treebank and the Danish Dependency Treebank.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"We are given a source string fJ 1 = f1:::fj :::fJ of length J, which is to be translated into a target string eI 1 = e1:::ei:::eI of length I. Among all possible target strings, we will choose the string with the highest probability: ^eI 1 = arg max eI 1 fPr(eI 1jfJ 1 )g = arg max eI 1 fPr(eI 1) Pr(fJ 1 jeI 1)g : (1) The argmax operation denotes the search problem, i.e. the generation of the output sentence in the target language.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The CFLex and CFNet knowledge sources provide positive evidence that a candidate NP and anaphor might be coreferent.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"This paper, however, provides a comprehensive overview of the data collection effort and its current state.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"If one system is better in 95% of the sample sets, we conclude that its higher BLEU score is statistically significantly better.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"See Section 5.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Reiche\u00e2\u0080\u0099s colleagues will make sure that the concept is waterproof.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Roughly speaking, previous work can be divided into three categories, namely purely statistical approaches, purely lexi\u00c2\u00ad cal rule-based approaches, and approaches that combine lexical information with sta\u00c2\u00ad tistical information.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"The focus of this work is on building POS taggers for foreign languages, assuming that we have an English POS tagger and some parallel text between the two languages.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Since pronouns carry little semantics of their own, resolving them depends almost entirely on context.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Finally we show the combining techniques degrade very little when a poor parser is added to the set.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"For example, the two NEs \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d and \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d have the following contexts.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Then, the remaining texts were annotated and cross-validated, always with discussions among the annotators.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"This is equivalent to the assumption used in probability estimation for na\u00efve Bayes classifiers, namely that the attribute values are conditionally independent when the target value is given.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":".","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This latter evaluation compares the performance of the system with that of several human judges since, as we shall show, even people do not agree on a single correct way to segment a text.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Mikheev et al.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"TAG's can be used to give the structural descriptions discussed by Gazdar (1985) for the unbounded nested dependencies in Norwedish, for cross serial dependencies in Dutch subordinate clauses, and for the nestings of paired English complementizers.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"By design, they readily capture regularities at the token-level.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"However, it is robust, efficient, and easy to implement.4 To perform the maximization in (7), we used the popular L-BFGS algorithm (Liu and Nocedal, 1989), which requires gradient information.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Another attempt at using global information can be found in (Borthwick, 1999).","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"These two restrictions impose the constraint that the result of composing any two structures should be a structure whose &quot;size&quot; is the sum of its constituents plus some constant For example, the operation 4, discussed in the case of CFG's (in Section 4.1) adds the constant equal to the sum of the length of the strings VI, un+r\u2022 Since we are considering formalisms with arbitrary structures it is difficult to precisely specify all of the restrictions on the composition operations that we believe would appropriately generalize the concatenation operation for the particular structures used by the formalism.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Since the inclusion of out-ofdomain test data was a very late decision, the participants were not informed of this.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"The type-level tag assignments T generate features associated with word types W . The tag assignments constrain the HMM emission parameters \u00ce\u00b8.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Nu mb er filters candidate if number doesn\u00e2\u0080\u0099t agree.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"In our experiment, we set the threshold of the TF\/ITF score empirically using a small development corpus; a finer adjustment of the threshold could reduce the number of such keywords.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For each pair of judges, consider one judge as the standard,.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The confidence intervals are computed by bootstrap resampling for BLEU, and by standard significance testing for the manual scores, as described earlier in the paper.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"So, we set a threshold that at least two examples are required to build a link.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"3.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"Formally, we define dependency graphs as follows: 3.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In a few cases, the criteria for correctness are made more explicit.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Each production can push or pop symbols on the stack as can be seen in the following productions that generate tree of the form shown in Figure 4b.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"While the paper mentioned a sorted variant, code was never released.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Our analysis and comparison focuses primarily on the one-to-one accuracy since it is a stricter metric than many-to-one accuracy, but also report many-to-one for completeness.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Analysis of the data revealed that the contextual role knowledge is especially helpful for resolving pronouns because, in general, they are semantically weaker than definite NPs.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"With regard to purely morphological phenomena, certain processes are not han\u00c2\u00ad dled elegantly within the current framework Any process involving reduplication, for instance, does not lend itself to modeling by finite-state techniques, since there is no way that finite-state networks can directly implement the copying operations required.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The edge from the root to the subtree for the derivation of 7i is labeled by the address ni.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"So, there is a limitation that IE can only be performed for a predefined task, like \u00e2\u0080\u009ccorporate mergers\u00e2\u0080\u009d or \u00e2\u0080\u009cmanagement succession\u00e2\u0080\u009d.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Subsets of partial hypotheses with coverage sets C of increasing cardinality c are processed.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We can use the preceding linguistic and annotation insights to build a manually annotated Arabic grammar in the manner of Klein and Manning (2003).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For novel texts, no lexicon that consists simply of a list of word entries will ever be entirely satisfactory, since the list will inevitably omit many constructions that should be considered words.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Although these existential NPs do not need a prior referent, they may occur multiple times in a document.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"(2010), we adopt a simpler na\u00c2\u00a8\u00c4\u00b1ve Bayes strategy, where all features are emitted independently.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The use of the Good-Turing equation presumes suitable estimates of the unknown expectations it requires.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The linear and nonerasing assumptions about the operations discussed in Section 4.1 require that each z, and yk is used exactly once to define the strings zi, ,z1,3.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"\u00e2\u0080\u00a2 Bridging links: the annotator is asked to specify the type as part-whole, cause-effect (e.g., She had an accident.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"However, their inverted variant implements a reverse trie using less CPU and the same amount of memory7.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"(7) \u03c6 s,t This is a somewhat less direct objective than used by Matsoukas et al, who make an iterative approximation to expected TER.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Word type N % Dic tion ary entr ies 2 , 5 4 3 9 7 . 4 7 Mor pho logi call y deri ved wor ds 3 0 . 1 1 Fore ign tran slite rati ons 9 0 . 3 4 Per son al na mes 5 4 2 . 0 7 cases.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(3)) to be defined over unlabeled as well as labeled instances.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"All experiments use ATB parts 1\u00e2\u0080\u00933 divided according to the canonical split suggested by Chiang et al.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Upon identifying an anaphoric expression (currently restricted to: pronouns, prepositional adverbs, definite noun phrases), the an- notator first marks the antecedent expression (currently restricted to: various kinds of noun phrases, prepositional phrases, verb phrases, sentences) and then establishes the link between the two.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"We will also directly compare with a baseline similar to the Matsoukas et al approach in order to measure the benefit from weighting phrase pairs (or ngrams) rather than full sentences.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The algorithm builds two classifiers iteratively: each iteration involves minimization of a continuously differential function which bounds the number of examples on which the two classifiers disagree.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The terrorism examples reflect fairly obvious relationships: people who are murdered are killed; agents that \u00e2\u0080\u009creport\u00e2\u0080\u009d things also \u00e2\u0080\u009cadd\u00e2\u0080\u009d and \u00e2\u0080\u009cstate\u00e2\u0080\u009d things; crimes that are \u00e2\u0080\u009cperpetrated\u00e2\u0080\u009d are often later \u00e2\u0080\u009ccondemned\u00e2\u0080\u009d.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"(2008) reported agreement between the teams (measured with Evalb) at 93.8% F1, the level of the CTB.","label":0},{"source_text":"There is no global pruning.","target_text":"The following two error criteria are used in our experiments: mWER: multi-reference WER: We use the Levenshtein distance between the automatic translation and several reference translations as a measure of the translation errors.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"We use two different similarity functions to define the edge weights among the foreign vertices and between vertices from different languages.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Again, this deserves further investigation.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The relevant variables are the set of token-level tags that appear before and after each instance of the ith word type; we denote these context pairs with the set {(tb, ta)} and they are contained in t(\u00e2\u0088\u0092i).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Performance improvements transfer to the Moses (Koehn et al., 2007), cdec (Dyer et al., 2010), and Joshua (Li et al., 2009) translation systems where our code has been integrated.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Once again we present both a non-parametric and a parametric technique for this task.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"The choice of the genre commentary resulted from the fact that an investigation of rhetorical structure, its interaction with other aspects of discourse structure, and the prospects for its automatic derivation are the key motivations for building up the corpus.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"This intuition is born out by the experimental results.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The following algorithm was then used to induce new rules: Let Count' (x) be the number of times feature x is seen with some known label in the training data.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"The features we used can be divided into 2 classes: local and global.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"both MENE and IdentiFinder used more training data than we did (we used only the official MUC 6 and MUC7 training data).","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"I = 1X21 N and N is a &quot;medium&quot; sized number so that it is feasible to collect 0(N) unlabeled examples.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Memory mapping also allows the same model to be shared across processes on the same machine.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Note that in line 4 the last visited position for the successor hypothesis must be m. Otherwise , there will be four uncovered positions for the predecessor hypothesis violating the restriction.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We settled on contrastive evaluations of 5 system outputs for a single test sentence.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Reference resolution involves finding words that co-refer to the same entity.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"A spelling rule might be a simple look-up for the string (e.g., a rule that Honduras is a location) or a rule that looks at words within a string (e.g., a rule that any string containing Mr. is a person).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"A promising direction for future work is to explicitly model a distribution over tags for each word type.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"On each step CoBoost searches for a feature and a weight so as to minimize either 40 or 40.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"the number of permutations carried out for the word reordering is given.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In this specific case, as these two titles could fill the same column of an IE table, we regarded them as paraphrases for the evaluation.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"What ought to be developed now is an annotation tool that can make use of the format, allow for underspecified annotations and visualize them accordingly.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"So, who won the competition?","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"shows some keywords with their scores.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"This offers the well-known advantages for inter- changability, but it raises the question of how to query the corpus across levels of annotation.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"SRILM inefficiently stores 64-bit pointers.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Unbounded dependencies between branches are not possible in such a system.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"For verbs we add two features.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Not only is concatenation of these strings possible, but head wrapping can be used to split a string and wrap it around another string.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"On the first of these-the B set-our system had 64% recall and 86% precision; on the second-the C set-it had 33% recall and 19% precision.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"67 95.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"In the second scheme, Head+Path, we in addition modify the label of every arc along the lifting path from the syntactic to the linear head so that if the original label is p the new label is p\u2193.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The second modification is more important, and is discussed in the next section.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"49 99.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our results show that BABAR achieves good performance in both domains, and that the contextual role knowledge improves performance, especially on pronouns.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We concentrate on those sets.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The following three sections elaborate these different stages is more detail.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The first author is supported by a National Defense Science and Engineering Graduate (NDSEG) fellowship.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Table 8b shows that verbal nouns are the hardest pre-terminal categories to identify.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Turning now to (1), we have the similar problem that splitting.into.ma3 'horse' andlu4 'way' is more costly than retaining this as one word .ma3lu4 'road.'","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Again, we can compute average scores for all systems for the different language pairs (Figure 6).","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"This class of formalisms have the properties that their derivation trees are local sets, and manipulate objects, using a finite number of composition operations that use a finite number of symbols.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In (2a), we want to split the two morphemes since the correct analysis is that we have the adverb :1 cai2 'just,' the modal verb neng2 'be able' and the main verb R: Hke4fu2 'overcome'; the competing analysis is, of course, that we have the noun :1 cai2neng2 'talent,' followed by }'lijke4fu2 'overcome.'","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"37 84.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"For this example, since the sequence Even News Broadcasting Corp. only appears once in the document, its longest substring that occurs in the same document is News Broadcasting Corp. In this case, News has an additional feature of I begin set to 1, Broadcasting has an additional feature of I continue set to 1, and Corp. has an additional feature of I end set to 1.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"0 Figure 5 An example of affixation: the plural affix.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"In many cases these failures in recall would be fixed by having better estimates of the actual prob\u00c2\u00ad abilities of single-hanzi words, since our estimates are often inflated.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Furthermore, the combination of pruning and vertical markovization of the grammar outperforms the Oracle results reported by Cohen and Smith.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"As described in Sproat (1995), the Chinese segmenter presented here fits directly into the context of a broader finite-state model of text analysis for speech synthesis.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"85 82.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"IL+-1Proof: Assume a pair of crossing constituents appears in the output of the constituent voting technique using k parsers.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"I \u00e2\u0080\u00a2 JAPANS :rl4 .\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\)\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"o\u00c2\u00b7'\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\:J\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7 \u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7'\\; . '.:: ..........0 6.51 9.51 : jj \/ JAPANESE OCTOPUS 10\u00c2\u00b728i\u00c2\u00a3 :_nc HOW SAY f B :rl4 :il: :wen2 t '- \u00e2\u0080\u00a2 :zhang!","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"It is also worth pointing out a connection with Daum\u00b4e\u2019s (2007) work that splits each feature into domain-specific and general copies.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"This feature includes named entities, which the ATB marks with a flat NP node dominating an arbitrary number of NNP pre-terminal daughters (Figure 2).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Storing state therefore becomes a time-space tradeoff; for example, we store state with partial hypotheses in Moses but not with each phrase.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"It is the performance we could achieve if an omniscient observer told us which parser to pick for each of the sentences.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We present two algorithms.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"5 68.1 34.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"41.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"There are two key benefits of this model architecture.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"First, from a large corpus, we extract all the NE instance pairs.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"3) A tight coupling with the speech recognizer output.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The Danish Dependency Treebank (DDT) comprises about 100K words of text selected from the Danish PAROLE corpus, with annotation of primary and secondary dependencies (Kromann, 2003).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Our open-source (LGPL) implementation is also available for download as a standalone package with minimal (POSIX and g++) dependencies.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Across all languages, high performance can be attained by selecting a single tag per word type.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"\u00c2\u00a3 : _ADV: 5.88 If:!","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"In application settings, this may be a profitable strategy.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The natural baseline (baseline) outperforms the pure IN system only for EMEA\/EP fren.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The tee pumping lemma states that if there is tree, t = 22 t2t3, generated by a CFG G, whose height is more than a predetermined bound k, then all trees of the form ti tP3 for each i > 0 will also generated by G (as shown in Figure 9b).","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"These conaUses lossy compression. bThe 8-bit quantized variant returned incorrect probabilities as explained in Section 3.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Extract NE pair instances with contexts From the four years of newspaper corpus, we extracted 1.9 million pairs of NE instances.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Following the system devised under the Qing emperor Kang Xi, hanzi have traditionally been classified according to a set of approximately 200 semantic radicals; members of a radical class share a particular structural component, and often also share a common meaning (hence the term 'semantic').","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"For example, in predicting if a word belongs to a word class, is either true or false, and refers to the surrounding context: if = true, previous word = the otherwise The parameters are estimated by a procedure called Generalized Iterative Scaling (GIS) (Darroch and Ratcliff, 1972).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Indeed there are several open issues.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This highly effective approach is not directly applicable to the multinomial models used for core SMT components, which have no natural method for combining split features, so we rely on an instance-weighting approach (Jiang and Zhai, 2007) to downweight domain-specific examples in OUT.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"87 Table 7: Test set results.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"(Webber et al., 2003)).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Recently, statistical NERs have achieved results that are comparable to hand-coded systems.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"A named entity recognizer (NER) is useful in many NLP applications such as information extraction, question answering, etc. On its own, a NER can also provide users who are looking for person or organization names with quick information.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The edge weights between the foreign language trigrams are computed using a co-occurence based similarity function, designed to indicate how syntactically similar the middle words of the connected trigrams are (\u00a73.2).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"These universal POS categories not only facilitate the transfer of POS information from one language to another, but also relieve us from using controversial evaluation metrics,2 by establishing a direct correspondence between the induced hidden states in the foreign language and the observed English labels.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"SRILM inefficiently stores 64-bit pointers.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Various segmentation approaches were then compared with human performance: 1.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"In Section 2, we brie y review our approach to statistical machine translation.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Acronyms (ACRO): Words made up of all capitalized letters in the text zone will be stored as acronyms (e.g., IBM).","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Better Arabic Parsing: Baselines, Evaluations, and Analysis","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Finally we show the combining techniques degrade very little when a poor parser is added to the set.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"As you can see in the figure, the accuracy for the domain is quite high except for the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set, which contains various expressions representing different relationships for an IE application.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"SSER: subjective sentence error rate: For a more detailed analysis, the translations are judged by a human test person.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"There is a (costless) transition between the NC node and f,.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"In our coreference resolver, we define \u00ce\u00b8 to be the set of all candidate antecedents for an anaphor.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Better Arabic Parsing: Baselines, Evaluations, and Analysis","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"This is because in many cases, the use of hyphens can be considered to be optional (e.g., third-quarter or third quarter).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The method reported in this paper makes use solely of unigram probabilities, and is therefore a zeroeth-order model: the cost of a particular segmentation is estimated as the sum of the costs of the individual words in the segmentation.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The problem of &quot;noise&quot; items that do not fall into any of the three categories also needs to be addressed.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"This assumption, however, is not inherent to type-based tagging models.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"A sentence was withheld from section 22 because its extreme length was troublesome for a couple of the parsers.'","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Figure 2 shows a screenshot (which is of somewhat limited value, though, as color plays a major role in signalling the different statuses of the information).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"In the pinyin transliterations a dash(-) separates syllables that may be considered part of the same phonological word; spaces are used to separate plausible phonological words; and a plus sign (+) is used, where relevant, to indicate morpheme boundaries of interest.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"The type-level posterior term can be computed according to, P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2) \u00e2\u0088\u009d Note that each round of sampling Ti variables takes time proportional to the size of the corpus, as with the standard token-level HMM.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Roughly speaking, the new algorithm presented in this paper performs a similar search, but instead minimizes a bound on the number of (unlabeled) examples on which two classifiers disagree.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Three state-of-the-art statistical parsers are combined to produce more accurate parses, as well as new bounds on achievable Treebank parsing accuracy.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Here, we process only full-form words within the translation procedure.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This is not unreasonable given the application to phrase pairs from OUT, but it suggests that an interesting alternative might be to use a plain log-linear weighting function exp(Ei Aifi(s, t)), with outputs in [0, oo].","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"A corpus of German newspaper commentaries has been assembled at Potsdam University, and annotated with different linguistic information, to different degrees.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"To see this, note thai the first two terms in the above equation correspond to the function that AdaBoost attempts to minimize in the standard supervised setting (Equ.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Indeed, as we shall show in Section 5, even human judges differ when presented with the task of segmenting a text into words, so a definition of the criteria used to determine that a given segmentation is correct is crucial before one can interpret such measures.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The gender, number, and scoping KSs eliminate candidates from consideration.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The Penn Arabic Treebank (ATB) syntactic guidelines (Maamouri et al., 2004) were purposefully borrowed without major modification from English (Marcus et al., 1993).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Our analysis and comparison focuses primarily on the one-to-one accuracy since it is a stricter metric than many-to-one accuracy, but also report many-to-one for completeness.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"As will be obvious later, their derivation tree sets will be local sets as are those of CFG's.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The model was parameterized such that the joint probability of a (label, feature-set) pair P(yi, xi) is written as The model assumes that (y, x) pairs are generated by an underlying process where the label is first chosen with some prior probability P(yi); the number of features mi is then chosen with some probability P(mi); finally the features are independently generated with probabilities P(xulyi).","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The alignment mapping is j ! i = aj from source position j to target position i = aj . The use of this alignment model raises major problems if a source word has to be aligned to several target words, e.g. when translating German compound nouns.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The binary language model from Section 5.2 and text phrase table were forced into disk cache before each run.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"(b) does the translation have the same meaning, including connotations?","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"First we present the non-parametric version of parser switching, similarity switching: The intuition for this technique is that we can measure a similarity between parses by counting the constituents they have in common.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"For each token, three types of features may be taken into account: the word form; the part-of-speech assigned by an automatic tagger; and labels on previously assigned dependency arcs involving the token \u2013 the arc from its head and the arcs to its leftmost and rightmost dependent, respectively.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The breakdown of the different types of words found by ST in the test corpus is given in Table 3.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Taking \/V, to be the number of examples an algorithm classified correctly (where all gold standard items labeled noise were counted as being incorrect), we calculated two measures of accuracy: See Tab.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Here, we conThis model is equivalent to the standard HMM ex cept that it enforces the one-word-per-tag constraint.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"This limits the number of NE category pairs to 2,000 and the number of NE pair instances to 0.63 million.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"In the interest of testing the robustness of these combining techniques, we added a fourth, simple nonlexicalized PCFG parser.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"This enables us to build a high performance NER without using separate classifiers to take care of global consistency or complex formulation on smoothing and backoff models (Bikel et al., 1997).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"It is generally considered to be fast (Pauls 29 \u2212 1 probabilities and 2' \u2212 2 non-zero backoffs. and Klein, 2011), with a default implementation based on hash tables within each trie node.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"These 140 NE categories are designed by extending MUC\u00e2\u0080\u0099s 7 NE categories with finer sub-categories (such as Company, Institute, and Political Party for Organization; and Country, Province, and City for Location) and adding some new types of NE categories (Position Title, Product, Event, and Natural Object).","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Pseudo-Projective Dependency Parsing","label":1},{"source_text":"There is no global pruning.","target_text":"We can do that . IbmS: Yes, wonderful.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Turning now to (1), we have the similar problem that splitting.into.ma3 'horse' andlu4 'way' is more costly than retaining this as one word .ma3lu4 'road.'","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In the denomi 11 We have two such lists, one containing about 17,000 full names, and another containing frequencies of.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Supervised methods have been applied quite successfully to the full MUC named-entity task (Bikel et al. 97).","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The difference in precision between similarity and Bayes switching techniques is significant, but the difference in recall is not.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"This is because our corpus is not annotated, and hence does not distinguish between the various words represented by homographs, such as, which could be \/adv jiangl 'be about to' orInc jiang4 '(military) general'-as in 1j\\xiao3jiang4 'little general.'","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"For example, take ..., says Maury Cooper, a vice president at S.&P.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Figure 5 shows how this model is implemented as part of the dictionary WFST.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"For the TM, this is: where cI(s, t) is the count in the IN phrase table of pair (s, t), po(s|t) is its probability under the OUT TM, and cI(t) = &quot;s, cI(s', t).","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The local feature groups are: Non-Contextual Feature: This feature is set to 1 for all tokens.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Applications such as machine translation use language model probability as a feature to assist in choosing between hypotheses.","label":0},{"source_text":"There is no global pruning.","target_text":"The computing time, the number of search errors, and the multi-reference WER (mWER) are shown as a function of t0.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"More examples are shown in Figure 5.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We carried out translation experiments in two different settings.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Recently, statistical NERs have achieved results that are comparable to hand-coded systems.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Most computational models for coreference resolution rely on properties of the anaphor and candidate antecedent, such as lexical matching, grammatical and syntactic features, semantic agreement, and positional information.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"For IE, the system must be able to distinguish between semantically similar noun phrases that play different roles in an event.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Reiche\u00e2\u0080\u0099s colleagues will make sure that the concept is waterproof.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"To combat the proliferation of parsing edges, we prune the lattices according to a hand-constructed lexicon of 31 clitics listed in the ATB annotation guidelines (Maamouri et al., 2009a).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"He used an additional maximum entropy classifier that tries to correct mistakes by using reference resolution.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We evaluate our approach on eight European languages (\u00a76), and show that both our contributions provide consistent and statistically significant improvements.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, some caveats are in order in comparing this method (or any method) with other approaches to seg\u00c2\u00ad mentation reported in the literature.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"AdaBoost.MH can be applied to the problem using these pseudolabels in place of supervised examples.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Lilo CFG's, TAG's, and HG's the derivation tree set of a MCTAG will be a local set.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"of Tokens No.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"Following the setup of Johnson (2007), we use the whole of the Penn Treebank corpus for training and evaluation on English.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Daum\u00b4e (2007) applies a related idea in a simpler way, by splitting features into general and domain-specific versions.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"An edge indicates that the two features must have the same label.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"If two systems\u2019 scores are close, this may simply be a random effect in the test data.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Set the decision list to include all rules whose (smoothed) strength is above some threshold Pmin.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The last affix in the list is the nominal plural f, men0.20 In the table are the (typical) classes of words to which the affix attaches, the number found in the test corpus by the method, the number correct (with a precision measure), and the number missed (with a recall measure).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"One possibility is to use n-grams based on mutual information.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Such analyzers propose multiple segmentation possibilities and their corresponding analyses for a token in isolation but have no means to determine the most likely ones.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"However, there is again local grammatical information that should favor the split in the case of (1a): both .ma3 'horse' and .ma3 lu4 are nouns, but only .ma3 is consistent with the classifier pil, the classifier for horses.21 By a similar argument, the preference for not splitting , lm could be strengthened in (lb) by the observation that the classifier 1'1* tiao2 is consistent with long or winding objects like , lm ma3lu4 'road' but not with,ma3 'horse.'","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"12 For English, our Evalb implementation is identical to the most recent reference (EVALB20080701).","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Storing state therefore becomes a time-space tradeoff; for example, we store state with partial hypotheses in Moses but not with each phrase.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"10) and trained both EM and L-BFGS for 1000 iterations.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Borth- wick (1999) successfully made use of other hand- coded systems as input for his MENE system, and achieved excellent results.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(b) POS tagging accuracy is lowest for maSdar verbal nouns (VBG,VN) and adjectives (e.g., JJ).","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Lexicon Feature: The string of the token is used as a feature.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The dependency graph in Figure 1 satisfies all the defining conditions above, but it fails to satisfy the condition ofprojectivity (Kahane et al., 1998): The arc connecting the head jedna (one) to the dependent Z (out-of) spans the token je (is), which is not dominated by jedna.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"More complex approaches such as the relaxation technique have been applied to this problem Fan and Tsai (1988}.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"While processing the source sentence monotonically, the initial state I is entered whenever there are no uncovered positions to the left of the rightmost covered position.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The links can solve the problem.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Because there might be some controversy about the exact definitions of such universals, this set of coarse-grained POS categories is defined operationally, by collapsing language (or treebank) specific distinctions to a set of categories that exists across all languages.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The alignment mapping is j ! i = aj from source position j to target position i = aj . The use of this alignment model raises major problems if a source word has to be aligned to several target words, e.g. when translating German compound nouns.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Using these states, we define partial hypothesis extensions, which are of the following type: (S0;C n fjg; j0) !","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"If the expression is longer or complicated (like \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d and \u00e2\u0080\u009cA\u00e2\u0080\u0099s purchase of B\u00e2\u0080\u009d), it is called \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d, i.e. a set of phrases which express the same thing or event.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Memory usage in PROBING is high, though SRILM is even larger, so where memory is of concern we recommend using TRIE, if it fits in memory.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The approach builds from an initial seed set for a category, and is quite similar to the decision list approach described in (Yarowsky 95).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Finally, we show that in application settings, the absence of gold segmentation lowers parsing performance by 2\u00e2\u0080\u00935% F1.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Matsoukas et al (2009) generalize it by learning weights on sentence pairs that are used when estimating relative-frequency phrase-pair probabilities.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"We collected around 300\u2013400 judgements per judgement type (adequacy or fluency), per system, per language pair.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The probabilities are incorporated into the DempsterShafer model using Equation 1.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The human judges were presented with the following definition of adequacy and fluency, but no additional instructions:","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The addition of vertical markovization enables non-pruned models to outperform all previously reported re12Cohen and Smith (2007) make use of a parameter (\u03b1) which is tuned separately for each of the tasks.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Blum and Mitchell 98) go on to give PAC results for learning in the cotraining case.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"(2010) and the posterior regular- ization HMM of Grac\u00c2\u00b8a et al.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Sentence (2) and (3) help to disambiguate one way or the other.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"It is. based on the traditional character set rather than the simplified character set used in Singapore and Mainland China.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"07 95.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"To prevent this we &quot;smooth&quot; the confidence by adding a small value, e, to both W+ and W_, giving at = Plugging the value of at from Equ.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"85 82.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Furthermore, the combination of pruning and vertical markovization of the grammar outperforms the Oracle results reported by Cohen and Smith.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Nonetheless, the prospect of a network of annotated discourse resources seems particularly promising if not only a single annotation layer is used but a whole variety of them, so that a systematic search for correlations between them becomes possible, which in turn can lead to more explanatory models of discourse structure.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"wo rd => na m e 2.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The first concerns how to deal with ambiguities in segmentation.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"2.1 Inverted Alignments.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"However, it is unlikely that other occurrences of News Broadcasting Corp. in the same document also co-occur with Even.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The POS distributions over the foreign trigram types are used as features to learn a better unsupervised POS tagger (\u00a75).","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We are especially grateful to Taylor Berg- Kirkpatrick for running additional experiments.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Supervised part-of-speech (POS) taggers, for example, approach the level of inter-annotator agreement (Shen et al., 2007, 97.3% accuracy for English).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The Leaf Ancestor metric measures the cost of transforming guess trees to the reference (Sampson and Babarczy, 2003).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"A different notion of information structure, is used in work such as that of (?), who tried to characterize felicitous constituent ordering (theme choice, in particular) that leads to texts presenting information in a natural, \u00e2\u0080\u009cflowing\u00e2\u0080\u009d way rather than with abrupt shifts of attention.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The reference medicine for Silapo is EPREX\/ERYPO, which contains epoetin alfa.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"The original OUT counts co(s, t) are weighted by a logistic function w\u03bb(s, t): To motivate weighting joint OUT counts as in (6), we begin with the \u201cideal\u201d objective for setting multinomial phrase probabilities 0 = {p(s|t), dst}, which is the likelihood with respect to the true IN distribution pi(s, t).","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"A crucial difference is that the number of parameters is greatly reduced as is the number of variables that are sampled during each iteration.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"i..f,..","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"There has also been some work on adapting the word alignment model prior to phrase extraction (Civera and Juan, 2007; Wu et al., 2005), and on dynamically choosing a dev set (Xu et al., 2007).","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"In this case nonalpha is the string formed by removing all upper\/lower case letters from the spelling (e.g., for Thomas E. Petry nonalpha= .","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"As the two NE categories are the same, we can\u00e2\u0080\u0099t differentiate phrases with different orders of par ticipants \u00e2\u0080\u0093 whether the buying company or the to-be-bought company comes first.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Saving state allows our code to walk the data structure exactly once per query.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"As we shall see, most of the linked sets are paraphrases.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Each learner is free to pick the labels for these instances.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"In other words, the set of hidden states F was chosen to be the fine set of treebank tags.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Furthermore, the systematic way in which particles are prefixed to one another and onto an open-class category gives rise to a distinct sort of morphological ambiguity: space-delimited tokens may be ambiguous between several different segmentation possibilities.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"For example, hanzi containing the INSECT radical !R tend to denote insects and other crawling animals; examples include tr wal 'frog,' feng1 'wasp,' and !Itt she2 'snake.'","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"We then evaluate the approach in two steps.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"For one thing, it is not clear who is to receive settlements or what should happen in case not enough teachers accept the offer of early retirement.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The problem can be represented as a graph with 2N vertices corresponding to the members of X1 and X2.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"0 55.3 34.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"Mai.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Given this similarity function, we define a nearest neighbor graph, where the edge weight for the n most similar vertices is set to the value of the similarity function and to 0 for all other vertices.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"This process produces a large set of caseframes coupled with a list of the noun phrases that they extracted.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"A different notion of information structure, is used in work such as that of (?), who tried to characterize felicitous constituent ordering (theme choice, in particular) that leads to texts presenting information in a natural, \u00e2\u0080\u009cflowing\u00e2\u0080\u009d way rather than with abrupt shifts of attention.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"For sorted lookup, we compare interpolation search, standard C++ binary search, and standard C++ set based on red-black trees.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Most of these groups follow a phrase-based statistical approach to machine translation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For the automatic evaluation, we used BLEU, since it is the most established metric in the field.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Lossy compressed models RandLM (Talbot and Osborne, 2007) and Sheffield (Guthrie and Hepple, 2010) offer better memory consumption at the expense of CPU and accuracy.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"36.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Once HMM parameters (\u00ce\u00b8, \u00cf\u0086) are drawn, a token-level tag and word sequence, (t, w), is generated in the standard HMM fashion: a tag sequence t is generated from \u00cf\u0086.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Besides information structure, the second main goal is to enhance current models of rhetorical structure.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"(7) \u03c6 s,t This is a somewhat less direct objective than used by Matsoukas et al, who make an iterative approximation to expected TER.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The key point is that the second constraint can be remarkably powerful in reducing the complexity of the learning problem.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"In (b) is a plausible segmentation for this sentence; in (c) is an implausible segmentation.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The most frequently occurring last words of organization names in cslist are compiled into a list of corporate suffixes, Corporate-Suffix- List.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Further, Maamouri and Bies (2004) argued that the English guidelines generalize well to other languages.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"We use two different similarity functions to define the edge weights among the foreign vertices and between vertices from different languages.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"To optimize left-to-right queries, we extend state to store backoff information: where m is the minimal context from Section 4.1 and b is the backoff penalty.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Ablation Analysis We evaluate the impact of incorporating various linguistic features into our model in Table 3.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"For unaligned words, we set the tag to the most frequent tag in the corresponding treebank.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"6. 3) all G o l d P O S 7 0 0.7 91 0.825 358 0.7 73 0.818 358 0.8 02 0.836 452 80.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Gabbard and Kulick (2008) show that there is significant attachment ambiguity associated with iDafa, which occurs in 84.3% of the trees in our development set.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"3 54.4 33.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Each feature group can be made up of many binary features.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The weak hypothesis chosen was then restricted to be a predictor in favor of this label.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"For the inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J), we drop the dependence on the target sentence length I. 2.2 Word Joining.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Besides the lack of a clear definition of what constitutes a correct segmentation for a given Chinese sentence, there is the more general issue that the test corpora used in these evaluations differ from system to system, so meaningful comparison between systems is rendered even more difficult.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"On the surface, our model may seem as a special case of Cohen and Smith in which \u03b1 = 0.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"On one end of the spectrum are clustering approaches that assign a single POS tag to each word type (Schutze, 1995; Lamar et al., 2010).","label":1},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"5 68.1 34.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Each xt E 2x is the set of features constituting the ith example.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"But Arabic contains a variety of linguistic phenomena unseen in English.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(Riloff and Jones 99) was brought to our attention as we were preparing the final version of this paper.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Other good classes include JADE and GOLD; other bad classes are DEATH and RAT.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Note, however, that there might be situations in which Zco in fact increases.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The addition of vertical markovization enables non-pruned models to outperform all previously reported re12Cohen and Smith (2007) make use of a parameter (\u03b1) which is tuned separately for each of the tasks.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Our full model (\u201cWith LP\u201d) outperforms the unsupervised baselines and the \u201cNo LP\u201d setting for all languages.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"With each iteration more examples are assigned labels by both classifiers, while a high level of agreement (> 94%) is maintained between them.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"For example, the two NEs \u00e2\u0080\u009cEastern Group Plc\u00e2\u0080\u009d and \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d have the following contexts.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The final score is obtained from: max e;e0 j2fJ\u00f4\u0080\u0080\u0080L;;Jg p($je; e0) Qe0 (e; I; f1; ; Jg; j); where p($je; e0) denotes the trigram language model, which predicts the sentence boundary $ at the end of the target sentence.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"3.1 Lexicon Component.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"In the last example, the less restrictive IbmS word reordering leads to a better translation, although the QmS translation is still acceptable.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The algorithm builds two classifiers in parallel from labeled and unlabeled data.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"68 96.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Because the information flow in our graph is asymmetric (from English to the foreign language), we use different types of vertices for each language.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"This alters generation of T as follows: n P (T |\u00cf\u0088) = n P (Ti|\u00cf\u0088) i=1 Note that this distribution captures the frequency of a tag across word types, as opposed to tokens.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In MUC6, the best result is achieved by SRA (Krupka, 1995).","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The first stage identifies a keyword in each phrase and joins phrases with the same keyword into sets.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"In addition to the Europarl test set, we also collected 29 editorials from the Project Syndicate website2, which are published in all the four languages of the shared task.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"During development, we sensed that the Recency and Syn- role KSs did not deserve to be on equal footing with the other KSs because their knowledge was so general.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The first two rows of the table are baselines.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"971,746 sentences of New York Times text were parsed using the parser of (Collins 96).1 Word sequences that met the following criteria were then extracted as named entity examples: whose head is a singular noun (tagged NN).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This implies, therefore, that a major factor in the performance of a Chinese segmenter is the quality of the base dictionary, and this is probably a more important factor-from the point of view of performance alone-than the particular computational methods used.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Interpolation search is therefore a form of binary search with better estimates informed by the uniform key distribution.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Pumping t2 will change only one branch and leave the other branch unaffected.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"selected; and that recall is defined to be the number of correct hits divided by the number of items that should have been selected.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"However, supervised methods rely on labeled training data, which is time-consuming and expensive to generate.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Recent work has made significant progress on unsupervised POS tagging (Me\u00c2\u00b4rialdo, 1994; Smith and Eisner, 2005; Haghighi and Klein, 2006; Johnson,2007; Goldwater and Griffiths, 2007; Gao and John son, 2008; Ravi and Knight, 2009).","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Figure 1: Translation of PCC sample commentary (STTS)2.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Case and Zone: If the token starts with a capital letter (initCaps), then an additional feature (init- Caps, zone) is set to 1.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"2) An improved language model, which takes into account syntactic structure, e.g. to ensure that a proper English verbgroup is generated.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"It is not easy to make a clear definition of \u00e2\u0080\u009cparaphrase\u00e2\u0080\u009d.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"pronunciations of individual words; they also need to compute intonational phrase boundaries in long utterances and assign relative prominence to words in those utterances.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"19 We note that it is not always clear in Wang, Li, and Chang's examples which segmented words.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This is appropriate in cases where it is sanctioned by Bayes\u2019 law, such as multiplying LM and TM probabilities, but for adaptation a more suitable framework is often a mixture model in which each event may be generated from some domain.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The probabilistic version of this procedure is straightforward: We once again assume independence among our various member parsers.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The first setting uses the European Medicines Agency (EMEA) corpus (Tiedemann, 2009) as IN, and the Europarl (EP) corpus (www.statmt.org\/europarl) as OUT, for English\/French translation in both directions.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Instead, we extend the variation n-gram method of Dickinson (2005) to compare annotation error rates in the WSJ and ATB.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"where the husband's family name is optionally prepended to the woman's full name; thus ;f:*lf#i xu3lin2-yan2hai3 would represent the name that Ms. Lin Yanhai would take if she married someone named Xu.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"One knowledge source, called WordSemCFSem, is analogous to CFLex: it checks whether the anaphor and candidate antecedent are substitutable for one another, but based on their semantic classes instead of the words themselves.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Along with IRSTLM and TPT, our binary format is memory mapped, meaning the file and in-memory representation are the same.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"However, their inverted variant implements a reverse trie using less CPU and the same amount of memory7.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"4 69.0 51.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The final block in table 2 shows models trained on feature subsets and on the SVM feature described in 3.4.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"In Equations 1 through 3 we develop the model for constructing our parse using na\u00efve Bayes classification.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(1992).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"In the terrorism domain, 1600 texts were used for training and the 40 test docu X \u00e2\u0088\u00a9Y =\u00e2\u0088\u0085 All sets of hypotheses (and their corresponding belief values) in the current model are crossed with the sets of hypotheses (and belief values) provided by the new evidence.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"A non-optimal analysis is shown with dotted lines in the bottom frame.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Here, an NE instance pair is any pair of NEs separated by at most 4 syntactic chunks; for example, \u00e2\u0080\u009cIBM plans to acquire Lotus\u00e2\u0080\u009d.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The sentence length probability p(JjI) is omitted without any loss in performance.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"And indeed, converging on annotation guidelines is even more difficult than it is with co-reference.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The orthographic normalization strategy we use is simple.10 In addition to removing all diacritics, we strip instances of taTweel J=J4.i, collapse variants of alif to bare alif,11 and map Ara bic punctuation characters to their Latin equivalents.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"So, 1: f, xue2shengl+men0 (student+PL) 'students' occurs and we estimate its cost at 11.43; similarly we estimate the cost of f, jiang4+men0 (general+PL) 'generals' (as in 'J' f, xiao3jiang4+men0 'little generals'), at 15.02.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"It can also be seen clearly in this plot that two of the Taiwan speakers cluster very closely together, and the third Tai\u00c2\u00ad wan speaker is also close in the most significant dimension (the x axis).","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Altun et al. (2005) proposed a technique that uses graph based similarity between labeled and unlabeled parts of structured data in a discriminative framework for semi-supervised learning.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Unfortunately, there is no standard corpus of Chinese texts, tagged with either single or multiple human judgments, with which one can compare performance of various methods.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In the namedentity problem each example is a (spelling,context) pair.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"f1; ; Jg denotes a coverage set including all positions from the starting position 1 to position J and j 2 fJ \u00f4\u0080\u0080\u0080L; ; Jg.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"4 69.0 51.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We can check, what the consequences of less manual annotation of results would have been: With half the number of manual judgements, we can distinguish about 40% of the systems, 10% less.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Hash tables are a common sparse mapping technique used by SRILM\u2019s default and BerkeleyLM\u2019s hashed variant.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In this paper we present a stochastic finite-state model for segmenting Chinese text into words, both words found in a (static) lexicon as well as words derived via the above-mentioned productive processes.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"A novel element of our model is the ability to capture type-level tag frequencies.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR uses a DempsterShafer decision model (Stefik, 1995) to combine the evidence provided by the knowledge sources.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Consequently, all three parsers prefer the nominal reading.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"\u00ce\u00b2 is the shared hyperparameter for the tag assignment prior and word feature multinomials.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Notice that the CC-domain is a special case.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"68 96.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Lexical and Morphological Ambiguity The rich morphological processes for deriving Hebrew stems give rise to a high degree of ambiguity for Hebrew space-delimited tokens.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"\u00e2\u0080\u009d, because Bush is the first word, the initial caps might be due to its position (as in \u00e2\u0080\u009cThey put a freeze on . . .","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Finally, other packages implement language model estimation while we are currently dependent on them to generate an ARPA file.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Email: gale@research.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Linear probing places at most one entry in each bucket.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In a fully supervised setting, the task is to learn a function f such that for all i = 1...m, f (xi,i, 12,i) = yz.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Roughly speaking, we say that a tee set contains trees with dependent paths if there are two paths p., = vim., and g., = in each 7 E r such that v., is some, possibly empty, shared initial subpath; v., and wi are not bounded in length; and there is some &quot;dependence&quot; (such as equal length) between the set of all v., and w., for each 7 Er.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"One hopes that such a corpus will be forth\u00c2\u00ad coming.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Extract NE pair instances with contexts From the four years of newspaper corpus, we extracted 1.9 million pairs of NE instances.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The independence of paths in the tree sets of the k tI grammatical formalism in this hierarchy can be shown by means of tree pumping lemma of the form t1ti3t .","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Presenting the output of several system allows the human judge to make more informed judgements, contrasting the quality of the different systems.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":" We evaluate the system's performance by comparing its segmentation 'Tudgments\" with the judgments of a pool of human segmenters, and the system is shown to perform quite well.","label":0},{"source_text":"Here we present two algorithms.","target_text":"From here on we will refer to the named-entity string itself as the spelling of the entity, and the contextual predicate as the context.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This significantly underperforms log-linear combination.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"29 \u00e2\u0080\u0094 95.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The quasi-monotone search performs best in terms of both error rates mWER and SSER.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Finally, the assignments of PoS tags to OOV segments is subject to language specific constraints relative to the token it was originated from.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Alternatively, h can be thought of as defining a decision list of rules x y ranked by their &quot;strength&quot; h(x, y).","label":0},{"source_text":"Here we present two algorithms.","target_text":"Several extensions of AdaBoost for multiclass problems have been suggested (Freund and Schapire 97; Schapire and Singer 98).","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"A sentence was withheld from section 22 because its extreme length was troublesome for a couple of the parsers.'","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Sparse lookup is a key subproblem of language model queries.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We define the following function: If Zco is small, then it follows that the two classifiers must have a low error rate on the labeled examples, and that they also must give the same label on a large number of unlabeled instances.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Crucially, the conventional orthographic form of MSA text is unvocalized, a property that results in a deficient graphical representation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"To this end, we picked 100 sentences at random containing 4,372 total hanzi from a test corpus.14 (There were 487 marks of punctuation in the test sentences, including the sentence-final periods, meaning that the average inter-punctuation distance was about 9 hanzi.)","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The Prague Dependency Treebank (PDT) consists of more than 1M words of newspaper text, annotated on three levels, the morphological, analytical and tectogrammatical levels (Haji\u02c7c, 1998).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Because of their size, the examples (Figures 2 to 4) appear at the end of the paper.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"This limits the number of NE category pairs to 2,000 and the number of NE pair instances to 0.63 million.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The numbers falling into the location, person, organization categories were 186, 289 and 402 respectively.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Only one of the first n positions which are not already aligned in a partial hypothesis may be chosen, where n is set to 4.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Their work used subject-verb, verb-object, and adjective-noun relations to compare the contexts surrounding an anaphor and candidate.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"When finished, the whole material is written into an XML-structured annotation file.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"To initialize the graph for label propagation we use a supervised English tagger to label the English side of the bitext.7 We then simply count the individual labels of the English tokens and normalize the counts to produce tag distributions over English word types.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We pick the parse that is most similar to the other parses by choosing the one with the highest sum of pairwise similarities.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"MENE (Maximum Entropy Named Entity) (Borth- wick, 1999) was combined with Proteus (a hand- coded system), and came in fourth among all MUC 7 participants.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"First, we evaluated BABAR using only the seven general knowledge sources.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"They are: 5We are grateful to an anonymous reviewer for pointing this out.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"For instance, in the recent IWSLT evaluation, first fluency annotations were solicited (while withholding the source sentence), and then adequacy annotations.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Section 2.1 describes how BABAR generates training examples to use in the learning process.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Methods that allow multiple segmentations must provide criteria for choosing the best segmentation.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The authors acknowledge the support of the NSF (CAREER grant IIS0448168, and grant IIS 0904684).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Their results are then compared with the results of an automatic segmenter.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"An example is in (i), where the system fails to group t;,f;?\"$?t!: lin2yang2gang3 as a name, because all three hanzi can in principle be separate words (t;,f; lin2 'wood';?\"$ yang2 'ocean'; ?t!; gang3 'harbor').","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"SRILM inefficiently stores 64-bit pointers.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"As an indication, in our core corpus, we found an average sentence length of 15.8 words and 1.8 verbs per sentence, whereas a randomly taken sample of ten commentaries from the national papers Su\u00c2\u00a8ddeutsche Zeitung and Frankfurter Allgemeine has 19.6 words and 2.1 verbs per sentence.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This is a straightforward technique that is arguably better suited to the adaptation task than the standard method of treating representative IN sentences as queries, then pooling the match results.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The first m pairs have labels yi, whereas for i = m + 1, , n the pairs are unlabeled.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"When keys are longer than 64 bits, we conserve space by replacing the keys with their 64-bit hashes.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Following the convention presented in earlier sections, we assume that each example is an instance pair of the from (xi ,i, x2,) where xj,, E 2x3 , j E 2}.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The two knowledge sources that use semantic expectations, WordSemCFSem and CFSemCFSem, always return values of -1 or 0.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"In contrast to the Bayesian HMM, \u00ce\u00b8t is not drawn from a distribution which has support for each of the n word types.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We evaluate our approach on seven languages: English, Danish, Dutch, German, Portuguese, Spanish, and Swedish.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"When a company buys another company, a paying event can occur, but these two phrases do not indicate the same event.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"At first glance, this seems only peripherally related to our work, since the specific\/general distinction is made for features rather than instances.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"16 As one reviewer points out, one problem with the unigram model chosen here is that there is still a. tendency to pick a segmentation containing fewer words.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"We also add an annotation for one-level iDafa (oneLevelIdafa) constructs since they make up more than 75% of the iDafa NPs in the ATB (Gabbard and Kulick, 2008).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The NIST06 and NIST08 evaluation sets were used for testing.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"97 78.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The difference in performance between pronouns and definite noun phrases surprised us.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For example, the company may refer to Company X in one paragraph and Company Y in another.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Figure 1 provides some statistics about this corpus.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"When OUT is large and distinct, its contribution can be controlled by training separate IN and OUT models, and weighting their combination.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We represent all morphological analyses of a given utterance using a lattice structure.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Evaluation We use 8 different measures to evaluate the performance of our system on the joint disambiguation task.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"Ties are rare in Bayes switching because the models are fine-grained \u2014 many estimated probabilities are involved in each decision.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The set of candidate constituents comes from the union of all the constituents suggested by the member parsers.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Limitations There are several limitations in the methods.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"There is a (costless) transition between the NC node and f,.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Finally, we make some improvements to baseline approaches.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Our second point of comparison is with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We also thank ChaoHuang Chang, reviewers for the 1994 ACL conference, and four anonymous reviewers for Computational Linguistics for useful comments.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"MUC7 has also seen hybrids of statistical NERs and hand-coded systems (Mikheev et al., 1998; Borthwick, 1999), notably Mikheev' s system, which achieved the best performance of 93.39% on the official NE test data.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Feature selection is implemented using a feature cutoff: features seen less than a small count during training will not be used.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The logistic function, whose outputs are in [0, 1], forces pp(s, t) <_ po(s, t).","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"Parser 3, the most accurate parser, was chosen 71% of the time, and Parser 1, the least accurate parser was chosen 16% of the time.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"One striking example is the error reduction for Spanish, which reduces error by 36.5% and 24.7% for the best and median settings respectively.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"A k-tape ATM, M, has a read-only input tape and k read-write work tapes.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"La ng ua ge 1T W + P RI O R + F E A T S E ng lis h D a ni s h D u tc h G e r m a n P or tu g u e s e S p a ni s h S w e di s h 2 1.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"A receives a votes, and B receives b votes.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"These estimates are in turn combined linearly with relative-frequency estimates from an in-domain phrase table.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Therefore, the number of fine tags varied across languages for our experiments; however, one could as well have fixed the set of HMM states to be a constant across languages, and created one mapping to the universal POS tagset.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Experiments using data from the Prague Dependency Treebank show that the combined system can handle nonprojective constructions with a precision sufficient to yield a significant improvement in overall parsing accuracy.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Several such particles may be prefixed onto a single stem, in which case the affixation is subject to strict linear precedence constraints.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Even with vocalization, there are linguistic categories that are difficult to identify without semantic clues.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In particular, the decision to represent arguments in verb- initial clauses as VP internal makes VSO and VOS configurations difficult to distinguish.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Approaches differ in the algorithms used for scoring and selecting the best path, as well as in the amount of contextual information used in the scoring process.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"T e r r o r i s m Ca sef ra me Semantic Classes <a ge nt > ass ass ina ted group, human inv esti gat ion int o < N P> event exp lod ed out sid e < N P> building N a t u r a l D i s a s t e r s Ca sef ra me Semantic Classes <a ge nt > inv esti gat ing cau se group, human sur viv or of < N P> event, natphenom hit wit h < N P> attribute, natphenom Figure 3: Semantic Caseframe Expectations Figure 2: Lexical Caseframe Expectations To illustrate how lexical expectations are used, suppose we want to determine whether noun phrase X is the antecedent for noun phrase Y. If they are coreferent, then X and Y should be substitutable for one another in the story.4 Consider these sentences: (S1) Fred was killed by a masked man with a revolver.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"07 80.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"), which precludes a single universal approach to adaptation.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Finally, since non-projective constructions often involve long-distance dependencies, the problem is closely related to the recovery of empty categories and non-local dependencies in constituency-based parsing (Johnson, 2002; Dienes and Dubey, 2003; Jijkoun and de Rijke, 2004; Cahill et al., 2004; Levy and Manning, 2004; Campbell, 2004).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"By definition, each existential NP uniquely specifies an object or concept, so we can infer that all instances of the same existential NP are coreferent (e.g., \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d always refers to the same entity).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The question is how to normalize the probabilities in such a way that smaller groupings have a better shot at winning.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"A token that is allCaps will also be initCaps.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"For instance, on Spanish, the absolute gap on median performance is 10%.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"There are two key benefits of this model architecture.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The only way to handle such phenomena within the framework described here is simply to expand out the reduplicated forms beforehand, and incorporate the expanded forms into the lexical transducer.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"e0; e are the last two target words, C is a coverage set for the already covered source positions and j is the last position visited.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"More details on the parsing algorithm can be found in Nivre (2003).","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Skipped (K): The translation of up to one word may be postponed . Verb (V): The translation of up to two words may be anticipated.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The approach gains leverage from natural redundancy in the data: for many named-entity instances both the spelling of the name and the context in which it appears are sufficient to determine its type.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"This is not completely surprising, since all systems use very similar technology.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Tsarfaty and Sima\u2019an (2007) have reported state-of-the-art results on Hebrew unlexicalized parsing (74.41%) albeit assuming oracle morphological segmentation.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, for our purposes it is not sufficient to repre\u00c2\u00ad sent the morphological decomposition of, say, plural nouns: we also need an estimate of the cost of the resulting word.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"In this case nonalpha is the string formed by removing all upper\/lower case letters from the spelling (e.g., for Thomas E. Petry nonalpha= .","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"'Malaysia.'","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"With a human evaluation we also showed that ATB inter-annotator agreement remains low relative to the WSJ corpus.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Little attention, however, has been paid to the structural descriptions that these formalisms can assign to strings, i.e. their strong generative capacity.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Figure 2 shows a screenshot (which is of somewhat limited value, though, as color plays a major role in signalling the different statuses of the information).","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"The following three sections elaborate these different stages is more detail.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"shortest match at each point.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Often, two systems can not be distinguished with a confidence of over 95%, so there are ranked the same.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Chunking is not enough to find such relationships.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The availability of comparable corpora is limited, which is a significant limitation on the approach.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We retain segmentation markers\u00e2\u0080\u0094which are consistent only in the vocalized section of the treebank\u00e2\u0080\u0094to differentiate between e.g. \u00ef\u00bf\u00bd \u00e2\u0080\u009cthey\u00e2\u0080\u009d and \u00ef\u00bf\u00bd + \u00e2\u0080\u009ctheir.\u00e2\u0080\u009d Because we use the vocalized section, we must remove null pronoun markers.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"We carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The discovered paraphrases can be a big help to reduce human labor and create a more comprehensive pattern set.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The text type are editorials instead of speech transcripts.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"First, we describe how the caseframes are represented and learned.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"These are shown, with their associated costs, as follows: ABj nc 4.0 AB C\/jj 6.0 CD \/vb 5.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"In all cases, the key is collapsed to its 64-bit hash.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In this subsection, we will report the results of the experiment, in terms of the number of words, phrases or clusters.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"If an unlabeled vertex does not have a path to any labeled vertex, this term ensures that the converged marginal for this vertex will be uniform over all tags, allowing the middle word of such an unlabeled vertex to take on any of the possible tags.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"It was motivated by the observation that the (Yarowsky 95) algorithm added a very large number of rules in the first few iterations.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Because all English vertices are going to be labeled, we do not need to disambiguate them by embedding them in trigrams.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"At each point during the derivation, the parser has a choice between pushing the next input token onto the stack \u2013 with or without adding an arc from the token on top of the stack to the token pushed \u2013 and popping a token from the stack \u2013 with or without adding an arc from the next input token to the token popped.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Standard SMT systems have a hierarchical parameter structure: top-level log-linear weights are used to combine a small set of complex features, interpreted as log probabilities, many of which have their own internal parameters and objectives.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"There are two key benefits of this model architecture.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"and f,.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"The results clearly indicate increased variation in the ATB relative to the WSJ, but care should be taken in assessing the magnitude of the difference.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"We model each parse as the decisions made to create it, and model those decisions as independent events.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Two of the Mainlanders also cluster close together but, interestingly, not particularly close to the Taiwan speakers; the third Mainlander is much more similar to the Taiwan speakers.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"In order to evaluate and advance this approach, it helps to feed into the knowledge base data that is already enriched with some of the desired information \u00e2\u0080\u0094 as in PCC.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"At first glance, we quickly recognize that many systems are scored very similar, both in terms of manual judgement and BLEU.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Note also that the costs currently used in the system are actually string costs, rather than word costs.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The development of automatic scoring methods is an open field of research.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We will report the evaluation results in the next subsection.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"By contrast, when we turn to a comparison of the three encoding schemes it is hard to find any significant differences, and the overall impression is that it makes little or no difference which encoding scheme is used, as long as there is some indication of which words are assigned their linear head instead of their syntactic head by the projective parser.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"But it conflates the coordinating and discourse separator functions of wa (<..4.b \u00ef\u00bf\u00bd \u00ef\u00bf\u00bd) into one analysis: conjunction(Table 3).","label":0},{"source_text":"There is no global pruning.","target_text":"For a trigram language model, the partial hypotheses are of the form (e0; e; C; j).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Global context from the whole document is available and can be exploited in a natural manner with a maximum entropy classifier.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We suggest that in unlexicalized PCFGs the syntactic context may be explicitly modeled in the derivation probabilities.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(3) shows learning curves for CoBoost.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We evaluate our model on seven languages exhibiting substantial syntactic variation.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Here, the term frequency (TF) is the frequency of a word in the bag and the inverse term frequency (ITF) is the inverse of the log of the frequency in the entire corpus.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We can now add a new weak hypothesis 14 based on a feature in X1 with a confidence value al hl and atl are chosen to minimize the function We now define, for 1 <i <n, the following virtual distribution, As before, Ztl is a normalization constant.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"There is a sizable literature on Chinese word segmentation: recent reviews include Wang, Su, and Mo (1990) and Wu and Tseng (1993).","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"If e < b then the key is not found.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"To lower the barrier of entrance to the competition, we provided a complete baseline MT system, along with data resources.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"2 Chinese ?l* han4zi4 'Chinese character'; this is the same word as Japanese kanji..","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"To set \u03b2, we used the same criterion as for \u03b1, over a dev corpus: The MAP combination was used for TM probabilities only, in part due to a technical difficulty in formulating coherent counts when using standard LM smoothing techniques (Kneser and Ney, 1995).3 Motivated by information retrieval, a number of approaches choose \u201crelevant\u201d sentence pairs from OUT by matching individual source sentences from IN (Hildebrand et al., 2005; L\u00a8u et al., 2007), or individual target hypotheses (Zhao et al., 2004).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"This has led previous workers to adopt ad hoc linear weighting schemes (Finch and Sumita, 2008; Foster and Kuhn, 2007; L\u00a8u et al., 2007).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"BABAR uses information extraction patterns to identify contextual roles and creates four contextual role knowledge sources using unsupervised learning.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"5 67.3 55.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"For statistics on this test set, refer to Figure 1.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"More details on the memory-based prediction can be found in Nivre et al. (2004) and Nivre and Scholz (2004).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Intuitively, as suggested by the example in the introduction, this is the right granularity to capture domain effects.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The Input The set of analyses for a token is thus represented as a lattice in which every arc corresponds to a specific lexeme l, as shown in Figure 1.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"An input ABCD can be represented as an FSA as shown in Figure 2(b).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(For some recent corpus-based work on Chinese abbreviations, see Huang, Ahrens, and Chen [1993].)","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Next we show that the ATB is similar to other tree- banks in gross statistical terms, but that annotation consistency remains low relative to English (\u00c2\u00a73).","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"An alternate approximation to (8) would be to let w,\\(s, t) directly approximate p\u02c6I(s, t).","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"07 80.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"able at http:\/\/nlp.stanford.edu\/projects\/arabic.shtml.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"However, until such standards are universally adopted in evaluating Chinese segmenters, claims about performance in terms of simple measures like percent correct should be taken with a grain of salt; see, again, Wu and Fung (1994) for further arguments supporting this conclusion.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"A promising direction for future work is to explicitly model a distribution over tags for each word type.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Otherwise, label the training data with the combined spelling\/contextual decision list, then induce a final decision list from the labeled examples where all rules (regardless of strength) are added to the decision list.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Topicalization of NP subjects in SVO configurations causes confusion with VO (pro-drop).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"For example, a good classifier would identify Mrs. Frank as a person, Steptoe & Johnson as a company, and Honduras as a location.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This is demonstrated by average scores over all systems, in terms of BLEU, fluency and adequacy, as displayed in Figure 5.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"A step of an ATM consists of reading a symbol from each tape and optionally moving each head to the left or right one tape cell.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Figure 2 shows an excerpt of a sentence from the Italian test set and the tags assigned by four different models, as well as the gold tags.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"While the generative power of CG's is greater that of CFG's, it appears to be highly constrained.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"So, this was a surprise element due to practical reasons, not malice.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Table 1 briefly describes the seven syntactic heuristics used by BABAR to resolve noun phrases.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"In contrast, NNP (proper nouns) form a large portion of vocabulary.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Because conjunctions are elevated in the parse trees when they separate recursive constituents, we choose the right sister instead of the category of the next word.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"Our results outperform strong unsupervised baselines as well as approaches that rely on direct projections, and bridge the gap between purely supervised and unsupervised POS tagging models.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"For each domain, phrases which contain the same keyword are gathered to build a set of phrases (Step 3).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Systems that generally do worse than others will receive a negative one.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Much of the study of grammatical systems in computational linguistics has been focused on the weak generative capacity of grammatical formalism.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"In application settings, this may be a profitable strategy.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"64 76.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"In an experiment on automatic rhetorical parsing, the RST-annotations and PoS tags were used by (Reitter 2003) as a training corpus for statistical classification with Support Vector Machines.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"One is the accuracy within a set of phrases which share the same keyword; the other is the accuracy of links.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Participants were also provided with two sets of 2,000 sentences of parallel text to be used for system development and tuning.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"What ought to be developed now is an annotation tool that can make use of the format, allow for underspecified annotations and visualize them accordingly.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"We smooth Prf(p \u2014* (s, p)) for rare and OOV segments (s E l, l E L, s unseen) using a \u201cper-tag\u201d probability distribution over rare segments which we estimate using relative frequency estimates for once-occurring segments.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Eight out of the thirteen errors in the high frequency phrases in the CC-domain are the phrases in \u00e2\u0080\u009cagree\u00e2\u0080\u009d.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Unfortunately, modifying the model to account for these kind of dependencies is not at all straightforward.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Since different judges judged different systems (recall that judges were excluded to judge system output from their own institution), we normalized the scores.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The first value reports resident size after loading; the second is the gap between post-loading resident memory and peak virtual memory.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Recent work by Finkel and Manning (2009) which re-casts Daum\u00b4e\u2019s approach in a hierarchical MAP framework may be applicable to this problem.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We present several variations for the lexical component P (T , W |\u00cf\u0088), each adding more complex pa rameterizations.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"It was our hope that this competition, which included the manual and automatic evaluation of statistical systems and one rulebased commercial system, will give further insight into the relation between automatic and manual evaluation.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"In this subsection, we will report the results of the experiment, in terms of the number of words, phrases or clusters.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Mixing, smoothing, and instance-feature weights are learned at the same time using an efficient maximum-likelihood procedure that relies on only a small in-domain development corpus.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"We model po(s|t) using a MAP criterion over weighted phrase-pair counts: and from the similarity to (5), assuming y = 0, we see that w\u03bb(s, t) can be interpreted as approximating pf(s, t)\/po(s, t).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Whether a language even has orthographic words is largely dependent on the writing system used to represent the language (rather than the language itself); the notion \"orthographic word\" is not universal.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"The learned patterns are then normalized and applied to the corpus.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"(4) is minimized by setting Since a feature may be present in only a few examples, W_ can be in practice very small or even 0, leading to extreme confidence values.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"For illustration, an English translation of one of the commentaries is given in Figure 1.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"So the success of the algorithm may well be due to its success in maximizing the number of unlabeled examples on which the two decision lists agree.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"It then computes a normalized Levenshtein edit distance between the extracted chain and the reference.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Previous Work.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"For instance, by altering the emission distribution parameters, Johnson (2007) encourages the model to put most of the probability mass on few tags.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The inverted alignment probability p(bijbi\u00f4\u0080\u0080\u00801; I; J) and the lexicon probability p(fbi jei) are obtained by relative frequency estimates from the Viterbi alignment path after the final training iteration.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As can be seen, GR and this \"pared-down\" statistical method perform quite similarly, though the statistical method is still slightly better.16 AG clearly performs much less like humans than these methods, whereas the full statistical algorithm, including morphological derivatives and names, performs most closely to humans among the automatic methods.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"For IE, the system must be able to distinguish between semantically similar noun phrases that play different roles in an event.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The importance of this property becomes clear in contrasting theories underlying GPSG (Gazdar, Klein, Pulluna, and Sag, 1985), and GB (as described by Berwick, 1984) with those underlying LFG and FUG.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"For seven out of eight languages a threshold of 0.2 gave the best results for our final model, which indicates that for languages without any validation set, r = 0.2 can be used.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Annotation of syntactic structure for the core corpus has just begun.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Lack of correct reference translations was pointed out as a short-coming of our evaluation.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"If there is a frequent multi-word sequence in a domain, we could use it as a keyword candidate.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"summit Sharm (a) Al-Sheikh summit Sharm (b) DTNNP Al-Sheikh in a corpus position without a bracketing label, then we also add \u00e2\u0088\u0097n, NIL) to M. We call the set of unique n-grams with multiple labels in M the variation nuclei of C. Bracketing variation can result from either annotation errors or linguistic ambiguity.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"In fact, it is very difficult to maintain consistent standards, on what (say) an adequacy judgement of 3 means even for a specific language pair.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"At first glance, the problem seems quite complex: a large number of rules is needed to cover the domain, suggesting that a large number of labeled examples is required to train an accurate classifier.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"0 70.9 42.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"our full model yields 39.3% average error reduction across languages when compared to the basic configuration (1TW).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Tsarfaty (2006) was the first to demonstrate that fully automatic Hebrew parsing is feasible using the newly available 5000 sentences treebank.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"If a phrase does not contain any keywords, the phrase is discarded.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Various verbal (e.g., \u00ef\u00bf\u00bd, .::) and adjectival.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"We define the following function: If Zco is small, then it follows that the two classifiers must have a low error rate on the labeled examples, and that they also must give the same label on a large number of unlabeled instances.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"It is worth noting that, although nonprojective constructions are less frequent in DDT than in PDT, they seem to be more deeply nested, since only about 80% can be projectivized with a single lift, while almost 95% of the non-projective arcs in PDT only require a single lift.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"and \u00e2\u0080\u009cH\u00e2\u0080\u009d represents \u00e2\u0080\u009cHanson Plc\u00e2\u0080\u009d.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"A direct-mapped cache makes BerkeleyLM faster on repeated queries, but their fastest (scrolling) cached version is still slower than uncached PROBING, even on cache-friendly queries.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"The gender, number, and scoping KSs eliminate candidates from consideration.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Many hanzi have more than one pronunciation, where the correct.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"These are written to the state s(wn1) and returned so that they can be used for the following query.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"The extent to which this constraint is enforced varies greatly across existing methods.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"gaolgaolxing4xing4 'happily' In the particular form of A-not-A reduplication illustrated in (3a), the first syllable of the verb is copied, and the negative markerbu4 'not' is inserted between the copy and the full verb.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Such sequences are given additional features of A begin, A continue, or A end, and the acronym is given a feature A unique.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The core of Yarowsky's algorithm is as follows: where h is defined by the formula in equation 2, with counts restricted to training data examples that have been labeled in step 2.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"The work reported here is closely related to [Ha- segawa et al. 04].","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"In both cases the investigators were able to achieve significant improvements over the previous best tagging results.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"73 81.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"At the same time, the n-gram error rate is sensitive to samples with extreme n-gram counts.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"This decreases the statistical significance of our results compared to those studies.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"All systems (except for Systran, which was not tuned to Europarl) did considerably worse on outof-domain training data.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"In section 5, we then evaluate the entire parsing system by training and evaluating on data from the Prague Dependency Treebank.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Asking the annotator to also formulate the question is a way of arriving at more reproducible decisions.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Not surprisingly some semantic classes are better for names than others: in our corpora, many names are picked from the GRASS class but very few from the SICKNESS class.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"In MUC6 and MUC7, the named entity task is defined as finding the following classes of names: person, organization, location, date, time, money, and percent (Chinchor, 1998; Sundheim, 1995) Machine learning systems in MUC6 and MUC 7 achieved accuracy comparable to rule-based systems on the named entity task.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This significantly underperforms log-linear combination.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"At the very least, we are creating a data resource (the manual annotations) that may the basis of future research in evaluation metrics.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The statistical systems seem to still lag behind the commercial rule-based competition when translating into morphological rich languages, as demonstrated by the results for English-German and English-French.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"Experiments in two domains showed that the contextual role knowledge improved coreference performance, especially on pronouns.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"In (1) the sequencema3lu4 cannot be resolved locally, but depends instead upon broader context; similarly in (2), the sequence :::tcai2neng2 cannot be resolved locally: 1.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"A number of PCC commentaries will be read by professional news speakers and prosodic features be annotated, so that the various annotation layers can be set into correspondence with intonation patterns.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"4 Traditional Arabic linguistic theory treats both of these types as subcategories of noun \u00ef\u00bf\u00bd '.i . Figure 1: The Stanford parser (Klein and Manning, 2002) is unable to recover the verbal reading of the unvocalized surface form 0 an (Table 1).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"While there are other obstacles to completing this idea, we believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In Table 7 we give results for several evaluation metrics.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"By taking the ratio of matching n-grams to the total number of n-grams in the system output, we obtain the precision pn for each n-gram order n. These values for n-gram precision are combined into a BLEU score: The formula for the BLEU metric also includes a brevity penalty for too short output, which is based on the total number of words in the system output c and in the reference r. BLEU is sensitive to tokenization.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"We therefore also normalized judgements on a per-sentence basis.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"The form fmnh, for example, can be understood as the verb \u201clubricated\u201d, the possessed noun \u201cher oil\u201d, the adjective \u201cfat\u201d or the verb \u201cgot fat\u201d.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Each set is assigned two values: belief and plausibility.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Feature weights were set using Och\u2019s MERT algorithm (Och, 2003).","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We would like to thank Prof. Ralph Grish- man, Mr. Takaaki Hasegawa and Mr. Yusuke Shinyama for useful comments, discussion and evaluation.","label":0},{"source_text":"Here we present two algorithms.","target_text":"Unlabeled examples in the named-entity classification problem can reduce the need for supervision to a handful of seed rules.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"On the surface, our model may seem as a special case of Cohen and Smith in which \u03b1 = 0.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"10 Here we use the Good-Turing estimate (Baayen 1989; Church and Gale 1991), whereby the aggregate probability of previously unseen instances of a construction is estimated as ni\/N, where N is the total number of observed tokens and n1 is the number of types observed only once.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Tables 4 and 5 also show that putting all of the contextual role KSs in play at the same time produces the greatest performance gain.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"8 1 2.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Input: Ja , wunderbar . Konnen wir machen . MonS: Yes, wonderful.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"The effect of the pruning threshold t0 is shown in Table 5.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"\u00e2\u0080\u0094 similar results have been observed across multiple languages.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In contrast to the Bayesian HMM, \u00ce\u00b8t is not drawn from a distribution which has support for each of the n word types.","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"Gather phrases using keywords Next, we select a keyword for each phrase \u00e2\u0080\u0093 the top-ranked word based on the TF\/IDF metric.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"This token may further embed into a larger utterance, e.g., \u2018bcl hneim\u2019 (literally \u201cin-the-shadow the-pleasant\u201d, meaning roughly \u201cin the pleasant shadow\u201d) in which the dominated Noun is modified by a proceeding space-delimited adjective.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Conversely, we can use the full rhetorical tree from the annotations and tune the co-reference module.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"98 15.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"On the other hand, we are interested in the application of rhetorical analysis or \u00e2\u0080\u0098discourse parsing\u00e2\u0080\u0099 (3.2 and 3.3), in text generation (3.4), and in exploiting the corpus for the development of improved models of discourse structure (3.5).","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"At first glance, this seems only peripherally related to our work, since the specific\/general distinction is made for features rather than instances.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"By design, they readily capture regularities at the token-level.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"2.4 Underspecified rhetorical structure.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"This is especially true in the case of quotations\u00e2\u0080\u0094which are common in the ATB\u00e2\u0080\u0094where (1) will follow a verb like (2) (Figure 1).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The machine learning community has been in a similar situation and has studied the combination of multiple classifiers (Wolpert, 1992; Heath et al., 1996).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The high 1 tone of J1l would not normally neutralize in this fashion if it were functioning as a word on its own.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The difference between the featureless model (+PRIOR) and our full model (+FEATS) is 13.6% and 7.7% average error reduction on best and median settings respectively.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"This group of features attempts to capture such information.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Each unlabeled pair (x1,i, x2,i) is represented as an edge between nodes corresponding to x1,i and X2,i in the graph.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The compressed variant uses block compression and is rather slow as a result.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Table 6: Example Translations for the Verbmobil task.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Moreover, they are used as substantives much 2 Unlike machine translation, constituency parsing is not significantly affected by variable word order.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This is can not be the only explanation, since the discrepancy still holds, for instance, for out-of-domain French-English, where Systran receives among the best adequacy and fluency scores, but a worse BLEU score than all but one statistical system.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Along with locking and background kernel operations such as prefaulting, this explains why wall time is not one-eighth that of the single-threaded case. aLossy compression with the same weights. bLossy compression with retuned weights. the non-lossy options.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Graph construction does not require any labeled data, but makes use of two similarity functions.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The model is often further restricted so that each source word is assigned to exactly one target word (Brown et al., 1993; Ney et al., 2000).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Cluster phrases based on Links We now have a set of phrases which share a keyword.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Tsarfaty (2006) argues that for Semitic languages determining the correct morphological segmentation is dependent on syntactic context and shows that increasing information sharing between the morphological and the syntactic components leads to improved performance on the joint task.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"However, when grammatical relations like subject and object are evaluated, parsing performance drops considerably (Green et al., 2009).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Sc opi ng filters candidate if outside the anaphor\u00e2\u0080\u0099s scope.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"In these experiments, the input lacks segmentation markers, hence the slightly different dev set baseline than in Table 6.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Approaches differ in the algorithms used for scoring and selecting the best path, as well as in the amount of contextual information used in the scoring process.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"Thus, the language generated by a grammar of a LCFRS is semilinear.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"Table 3: Training and test conditions for the Verbmobil task (*number of words without punctuation marks).","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In Input: {(x1,i, Initialize: Vi, j : e(xi) = 0.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"This is true of the widely used link grammar parser for English (Sleator and Temperley, 1993), which uses a dependency grammar of sorts, the probabilistic dependency parser of Eisner (1996), and more recently proposed deterministic dependency parsers (Yamada and Matsumoto, 2003; Nivre et al., 2004).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If any of the tokens from to is in Person-Prefix- List, then another feature Person-Prefix is set to 1.","label":0},{"source_text":"Here we present two algorithms.","target_text":"A large number of rules is needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classi- However, we show that the use of data can reduce the requirements for supervision to just 7 simple &quot;seed&quot; rules.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"This is less than the 694 judgements 2004 DARPA\/NIST evaluation, or the 532 judgements in the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"yu2 'fish.'","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"5.1 Parsing Models.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We use w erations of sampling (see Figure 2 for a depiction).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The evaluation framework for the shared task is similar to the one used in last year\u2019s shared task.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Our System Wang, Li, and Chang a. 1\\!f!IP Eflltii \/1\\!f!J:P $1til I b. agm: I a m: c. 5 Bf is Bf 1 d. \"*:t: w _t ff 1 \"* :t: w_tff 1 g., , Transliteration\/Translation chen2zhongl-shenl qu3 'music by Chen Zhongshen ' huang2rong2 youlyoul de dao4 'Huang Rong said soberly' zhangl qun2 Zhang Qun xian4zhang3 you2qingl shang4ren2 hou4 'after the county president You Qing had assumed the position' lin2 quan2 'Lin Quan' wang2jian4 'Wang Jian' oulyang2-ke4 'Ouyang Ke' yinl qi2 bu4 ke2neng2 rong2xu3 tai2du2 er2 'because it cannot permit Taiwan Independence so' silfa3-yuan4zhang3 lin2yang2-gang3 'president of the Judicial Yuan, Lin Yanggang' lin2zhangl-hu2 jiangl zuo4 xian4chang3 jie3shuol 'Lin Zhanghu will give an ex\u00c2\u00ad planation live' jin4\/iang3 nian2 nei4 sa3 xia4 de jinlqian2 hui4 ting2zhi3 'in two years the distributed money will stop' gaoltangl da4chi2 ye1zi0 fen3 'chicken stock, a tablespoon of coconut flakes' you2qingl ru4zhu3 xian4fu3 lwu4 'after You Qing headed the county government' Table 5 Performance on morphological analysis.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"It is difficult when IN and OUT are dissimilar, as they are in the cases we study.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Unlike GCFG's, however, the composition operations of LCFRS's are restricted to be linear (do not duplicate unboundedly large structures) and nonerasing (do not erase unbounded structures, a restriction made in most modern transformational grammars).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The resulting structural differences between tree- banks can account for relative differences in parsing performance.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"For example, one parser could be more accurate at predicting noun phrases than the other parsers.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"63 95.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Supervised methods have been applied quite successfully to the full MUC named-entity task (Bikel et al. 97).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"This number must be less than or equal to n \u00f4\u0080\u0080\u0080 1.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The observed performance gains, coupled with the simplicity of model implementation, makes it a compelling alternative to existing more complex counterparts.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"In general, the neighborhoods can be more diverse and we allow a soft label distribution over the vertices.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"3.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"The set of candidate constituents comes from the union of all the constituents suggested by the member parsers.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"The number of judgements is additionally fragmented by our breakup of sentences into in-domain and out-of-domain.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"In Section 6 we dis\u00c2\u00ad cuss other issues relating to how higher-order language models could be incorporated into the model.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"1 74.5 56.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"Different annotations of the same text are mapped into the same data structure, so that search queries can be formulated across annotation levels.","label":0},{"source_text":"There is no global pruning.","target_text":"e0; e are the last two target words, C is a coverage set for the already covered source positions and j is the last position visited.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"For the manual scoring, we can distinguish only half of the systems, both in terms of fluency and adequacy.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The problem of &quot;noise&quot; items that do not fall into any of the three categories also needs to be addressed.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"The cost of storing these averages, in bits, is Because there are comparatively few unigrams, we elected to store them byte-aligned and unquantized, making every query faster.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"In (Reitter, Stede 2003) we went a different way and suggested URML5, an XML format for underspecifying rhetorical structure: a number of relations can be assigned instead of a single one, competing analyses can be represented with shared forests.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"It is natural that the larger the data in the domain, the more keywords are found.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"keywords Step 3 Sets of phrases based on keywords Step 4 Links between sets of phrases All the contexts collected for a given domain are gathered in a bag and the TF\/ITF scores are calculated for all the words except stopwords in the bag.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Only one of the first n positions which are not already aligned in a partial hypothesis may be chosen, where n is set to 4.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"logical rules, and personal names; the transitive closure of the resulting machine is then computed.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Model components cascade, so the row corresponding to +FEATS also includes the PRIOR component (see Section 3).","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"Similar behavior is observed when adding features.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The focus of this work is on building POS taggers for foreign languages, assuming that we have an English POS tagger and some parallel text between the two languages.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"The probability of the classes assigned to the words in a sentence in a document is defined as follows: where is determined by the maximum entropy classifier.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For the seen word ir, 'gen\u00c2\u00ad erals,' there is an c:NC transduction from to the node preceding ir,; this arc has cost cost( f,) - cost(unseen(f,)), so that the cost of the whole path is the desired cost( f,).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Evaluation results for links","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In the next section, we show how an ATM can accept the strings generated by a grammar in a LCFRS formalism in logspace, and hence show that each family can be recognized in polynomial time.","label":0},{"source_text":"Two general approaches are presented and two combination techniques are described for each approach.","target_text":"The development set contained 44088 constituents in 2416 sentences and the test set contained 30691 constituents in 1699 sentences.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"3.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"As illustrated by MCTAG's, it is possible for a formalism to give tree sets with bounded dependent paths while still sharing the constrained rewriting properties of CFG's, HG's, and TAG's.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"A less canonical representation of segmental morphology is triggered by a morpho-phonological process of omitting the definite article h when occurring after the particles b or l. This process triggers ambiguity as for the definiteness status of Nouns following these particles.We refer to such cases in which the concatenation of elements does not strictly correspond to the original surface form as super-segmental morphology.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Even when there is training data available in the domain of interest, there is often additional data from other domains that could in principle be used to improve performance.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Table 1: Syntactic Seeding Heuristics BABAR\u00e2\u0080\u0099s reliable case resolution heuristics produced a substantial set of anaphor\/antecedent resolutions that will be the training data used to learn contextual role knowledge.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Purely statistical approaches have not been very popular, and so far as we are aware earlier work by Sproat and Shih (1990) is the only published instance of such an approach.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"As the reviewer also points out, this is a problem that is shared by, e.g., probabilistic context-free parsers, which tend to pick trees with fewer nodes.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"For all variants, we found that BerkeleyLM always rounds the floating-point mantissa to 12 bits then stores indices to unique rounded floats.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"Due to the dramatic fiscal situation in Brandenburg she now surprisingly withdrew legislation drafted more than a year ago, and suggested to decide on it not before 2003.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"As can be seen in the example, the first two phrases have a different order of NE names from the last two, so we can determine that the last two phrases represent a reversed relation.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Many statistical or machine-learning approaches for natural language problems require a relatively large amount of supervision, in the form of labeled training examples.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Denote by g3(x) = Et crithl(x) , j E {1,2} the unthresholded strong-hypothesis (i.e., f3 (x) = sign(gi (x))).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"(Brandt 1996) extended these ideas toward a conception of kommunikative Gewichtung (\u00e2\u0080\u0098communicative-weight assignment\u00e2\u0080\u0099).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We compare the manually annotated grammar, which we incorporate into the Stanford parser, to both the Berkeley (Petrov et al., 2006) and Bikel (Bikel, 2004) parsers.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"As such, global information from the whole context of a document is important to more accurately recognize named entities.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"For instance: if 10 systems participate, and one system does better than 3 others, worse then 2, and is not significant different from the remaining 4, its rank is in the interval 3\u20137.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"One hopes that such a corpus will be forth\u00c2\u00ad coming.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"For example, one parser could be more accurate at predicting noun phrases than the other parsers.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The local features used are similar to those used in BBN' s IdentiFinder (Bikel et al., 1999) or MENE (Borthwick, 1999).","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"For example, a story can mention \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d, \u00e2\u0080\u009cthe White House\u00e2\u0080\u009d, or \u00e2\u0080\u009cthe weather\u00e2\u0080\u009d without any prior referent in the story.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The first four affixes are so-called resultative affixes: they denote some prop\u00c2\u00ad erty of the resultant state of a verb, as in E7 wang4bu4-liao3 (forget-not-attain) 'cannot forget.'","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The errors shown are from the Berkeley parser output, but they are representative of the other two parsing models.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective treebanks is often neglected because of the relative scarcity of problematic constructions.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Pr(eI 1) is the language model of the target language, whereas Pr(fJ 1 jeI1) is the transla tion model.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"More judgements would have enabled us to make better distinctions, but it is not clear what the upper limit is.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We now describe the CoBoost algorithm for the named entity problem.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"2.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The translation of one position in the source sentence may be postponed for up to L = 3 source positions, and the translation of up to two source positions may be anticipated for at most R = 10 source positions.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"na me =>1 ha nzi fa mi ly 2 ha nzi gi ve n 3.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"We evaluated the results based on two metrics.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Standard SMT systems have a hierarchical parameter structure: top-level log-linear weights are used to combine a small set of complex features, interpreted as log probabilities, many of which have their own internal parameters and objectives.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"A very small excerpt from an Italian-English graph is shown in Figure 1.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"The links can solve the problem.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Smith estimates Lotus will make a profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d, our system extracts \u00e2\u0080\u009cSmith esti mates Lotus\u00e2\u0080\u009d as an instance.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"SRILM\u2019s compact variant, IRSTLM, MITLM, and BerkeleyLM\u2019s sorted variant are all based on this technique.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"While the semantic aspect of radicals is by no means completely predictive, the semantic homogeneity of many classes is quite striking: for example 254 out of the 263 examples (97%) of the INSECT class listed by Wieger (1965, 77376) denote crawling or invertebrate animals; similarly 21 out of the 22 examples (95%) of the GHOST class (page 808) denote ghosts or spirits.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Figure 2 shows timing results.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"While automatic measures are an invaluable tool for the day-to-day development of machine translation systems, they are only a imperfect substitute for human assessment of translation quality, or as the acronym BLEU puts it, a bilingual evaluation understudy.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"We incur some additional memory cost due to storing state in each hypothesis, though this is minimal compared with the size of the model itself.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"For each terminal, the Leaf Ancestor metric extracts the shortest path to the root.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"We see from these results that the behavior of the parametric techniques are robust in the presence of a poor parser.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"(Again, the goal of also in structural features.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Coverage indicates the fraction of hypotheses in which the character yield exactly matched the reference.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"\u00e2\u0080\u00a2 Some tools would allow for the desired annotation mode, but are so complicated (they can be used for many other purposes as well) that annotators take a long time getting used to them.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(See Sproat and Shih 1995.)","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"For example, a story can mention \u00e2\u0080\u009cthe FBI\u00e2\u0080\u009d, \u00e2\u0080\u009cthe White House\u00e2\u0080\u009d, or \u00e2\u0080\u009cthe weather\u00e2\u0080\u009d without any prior referent in the story.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Finally we show the combining techniques degrade very little when a poor parser is added to the set.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The first unsupervised algorithm we describe is based on the decision list method from (Yarowsky 95).","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The difference in precision between similarity and Bayes switching techniques is significant, but the difference in recall is not.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"In both cases, SRILM walks its trie an additional time to minimize context as mentioned in Section 4.1.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"This means that the PCC cannot grow particularly quickly.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The Expectation Maximization (EM) algorithm (Dempster, Laird and Rubin 77) is a common approach for unsupervised training; in this section we describe its application to the named entity problem.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"29 \u00e2\u0080\u0094 95.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"If an unlabeled vertex does not have a path to any labeled vertex, this term ensures that the converged marginal for this vertex will be uniform over all tags, allowing the middle word of such an unlabeled vertex to take on any of the possible tags.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"Phrase tables were extracted from the IN and OUT training corpora (not the dev as was used for instance weighting models), and phrase pairs in the intersection of the IN and OUT phrase tables were used as positive examples, with two alternate definitions of negative examples: The classifier trained using the 2nd definition had higher accuracy on a development set.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Additionally, it works about 3 times as fast as the IBM style search.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"Lilo CFG's, TAG's, and HG's the derivation tree set of a MCTAG will be a local set.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"This data set of manual judgements should provide a fruitful resource for research on better automatic scoring methods.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"The local feature groups are: Non-Contextual Feature: This feature is set to 1 for all tokens.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The 1-bit sign is almost always negative and the 8-bit exponent is not fully used on the range of values, so in practice this corresponds to quantization ranging from 17 to 20 total bits.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Hash tables are a common sparse mapping technique used by SRILM\u2019s default and BerkeleyLM\u2019s hashed variant.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"While processing the source sentence monotonically, the initial state I is entered whenever there are no uncovered positions to the left of the rightmost covered position.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"35 76.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"A more forceful approach for encoding sparsity is posterior regularization, which constrains the posterior to have a small number of expected tag assignments (Grac\u00c2\u00b8a et al., 2009).","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"For RandLM and IRSTLM, the effect of caching can be seen on speed and memory usage.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"For even larger models, storing counts (Talbot and Osborne, 2007; Pauls and Klein, 2011; Guthrie and Hepple, 2010) is a possibility.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Simple Type-Level Unsupervised POS Tagging","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"In both cases the investigators were able to achieve significant improvements over the previous best tagging results.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"All language model queries issued by machine translation decoders follow a left-to-right pattern, starting with either the begin of sentence token or null context for mid-sentence fragments.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"All the links in the \u00e2\u0080\u009cCC-domain are shown in Step 4 in subsection 3.2.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"An examination of the subjects' bracketings confirmed that these instructions were satisfactory in yielding plausible word-sized units.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"Equ.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The new algorithm, which we call CoBoost, uses labeled and unlabeled data and builds two classifiers in parallel.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"We divide up each test set into blocks of 20 sentences (100 blocks for the in-domain test set, 53 blocks for the out-of-domain test set), check for each block, if one system has a higher BLEU score than the other, and then use the sign test.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This &quot;default&quot; feature type has 100% coverage (it is seen on every example) but a low, baseline precision.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"In turn we use two sorts of heuristics, orthogonal to one another, to prune segmentation possibilities based on lexical and grammatical constraints.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Floating point values may be stored in the trie exactly, using 31 bits for non-positive log probability and 32 bits for backoff5.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Our method does not assume any knowledge about the target language (in particular no tagging dictionary is assumed), making it applicable to a wide array of resource-poor languages.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Parser 3, the most accurate parser, was chosen 71% of the time, and Parser 1, the least accurate parser was chosen 16% of the time.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"In our experiments we set the parameter values randomly, and then ran EM to convergence.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Clearly, retaining the original frequencies is important for good performance, and globally smoothing the final weighted frequencies is crucial.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Finally, our U (unparsed) measure is used to report the number of sentences to which our system could not propose a joint analysis.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"Furthermore, by inverting the transducer so that it maps from phonemic transcriptions to hanzi sequences, one can apply the segmenter to other problems, such as speech recognition (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"A different but supplementary perspective on discourse-based information structure is taken 11ventionalized patterns (e.g., order of informa by one of our partner projects, which is inter tion in news reports).","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"The search starts in the hypothesis (I; f;g; 0).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For eight judges, ranging k between 1 and 8 corresponded to a precision score range of 90% to 30%, meaning that there were relatively few words (30% of those found by the automatic segmenter) on which all judges agreed, whereas most of the words found by the segmenter were such that one human judge agreed.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Obviously, the presence of a title after a potential name N increases the probability that N is in fact a name.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If the token is the first word of a sentence, then this feature is set to 1.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Projectivizing a dependency graph by lifting nonprojective arcs is a nondeterministic operation in the general case.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"So, who won the competition?","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"m(S) represents the belief that the correct hypothesis is included in S. The model assumes that evidence also arrives as a probability density function (pdf) over sets of hypotheses.6 Integrating new evidence into the existing model is therefore simply a matter of defining a function to merge pdfs, one representing the current belief system and one representing the beliefs of the new evidence.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"An alternate approximation to (8) would be to let w,\\(s, t) directly approximate p\u02c6I(s, t).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"To improve agreement during the revision process, a dual-blind evaluation was performed in which 10% of the data was annotated by independent teams.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"Discriminative Instance Weighting for Domain Adaptation in Statistical Machine Translation","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"The out-of-domain test set differs from the Europarl data in various ways.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"32 81.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Table 5 provides insight into the behavior of different models in terms of the tagging lexicon they generate.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(2003), which gives 96.8% accuracy on the test set.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"See Section 5.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"(a) Of the high frequency phrasal categories, ADJP and SBAR are the hardest to parse.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"At first glance, the problem seems quite complex: a large number of rules is needed to cover the domain, suggesting that a large number of labeled examples is required to train an accurate classifier.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Not every annotator was fluent in both the source and the target language.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"For example, suppose one is building a ITS system for Mandarin Chinese.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"A number of PCC commentaries will be read by professional news speakers and prosodic features be annotated, so that the various annotation layers can be set into correspondence with intonation patterns.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The Potsdam Commentary Corpus","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Following this method, we repeatedly \u2014 say, 1000 times \u2014 sample sets of sentences from the output of each system, measure their BLEU score, and use these 1000 BLEU scores as basis for estimating a confidence interval.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The graph was constructed using 2 million trigrams; we chose these by truncating the parallel datasets up to the number of sentence pairs that contained 2 million trigrams.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The resulting algorithm has a complexity of O(n!).","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Blum and Mitchell 98) give an example that illustrates just how powerful the second constraint can be.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The perplexity for the trigram language model used is 26:5.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The CoBoost algorithm just described is for the case where there are two labels: for the named entity task there are three labels, and in general it will be useful to generalize the CoBoost algorithm to the multiclass case.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"In addition, the restricted version of CG's (discussed in Section 6) generates tree sets with independent paths and we hope that it can be included in a more general definition of LCFRS's containing formalisms whose tree sets have path sets that are themselves LCFRL's (as in the case of the restricted indexed grammars, and the hierarchy defined by Weir).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Our model outperforms theirs on four out of five languages on the best hyperparameter setting and three out of five on the median setting, yielding an average absolute difference across languages of 12.9% and 3.9% for best and median settings respectively compared to their best EM or LBFGS performance.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The way we cant distinction between system performance.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"We proposed an unsupervised method to discover paraphrases from a large untagged corpus.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Systems that generally do better than others will receive a positive average normalizedjudgement per sentence.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"The verbal reading arises when the maSdar has an NP argument which, in vocalized text, is marked in the accusative case.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Due to many similarly performing systems, we are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(Blum and Mitchell 98) go on to give PAC results for learning in the cotraining case.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"(Brandt 1996) extended these ideas toward a conception of kommunikative Gewichtung (\u00e2\u0080\u0098communicative-weight assignment\u00e2\u0080\u0099).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"36.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Annotation consistency is important in any supervised learning task.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The approach builds from an initial seed set for a category, and is quite similar to the decision list approach described in (Yarowsky 95).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"basically complete, yet some improvements and extensions are still under way.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"The availability of these resources guided our selection of foreign languages.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"We assume that the goal in dependency parsing is to construct a labeled dependency graph of the kind depicted in Figure 1.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Most similar is scrolling queries, wherein left-to-right queries that add one word at a time are optimized.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"In addition, each feature function is a binary function.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Contains digits.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"When the maSdar lacks a determiner, the constituent as a whole resem bles the ubiquitous annexation construct \u00ef\u00bf\u00bd ?f iDafa.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"If one system is perfect, another has slight flaws and the third more flaws, a judge is inclined to hand out judgements of 5, 4, and 3.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"One side of the decision making process is when we choose to believe a constituent should be in the parse, even though only one parser suggests it.","label":0},{"source_text":"This paper talks about Pseudo-Projective Dependency Parsing.","target_text":"Using this encoding scheme, the arc from je to Z in Figure 2 would be assigned the label AuxP\u2191Sb (signifying an AuxP that has been lifted from a Sb).","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"This implies, therefore, that a major factor in the performance of a Chinese segmenter is the quality of the base dictionary, and this is probably a more important factor-from the point of view of performance alone-than the particular computational methods used.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Of course, we.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"The state-to-tag mapping is obtained from the best hyperparameter setting for 11 mapping shown in Table 3.","label":0},{"source_text":"There is no global pruning.","target_text":"No: Predecessor coverage set Successor coverage set 1 (f1; ;mg n flg ; l0) !","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"3.5 Improved models of discourse.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"Models that employ this strategy are denoted hsp.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"Hence, s < k. But by addition of the votes on the two parses, s > 2N-11> k, a contradiction.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"There is a (costless) transition between the NC node and f,.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"For inference, we are interested in the posterior probability over the latent variables in our model.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The method reported in this paper makes use solely of unigram probabilities, and is therefore a zeroeth-order model: the cost of a particular segmentation is estimated as the sum of the costs of the individual words in the segmentation.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(student+plural) 'students,' which is derived by the affixation of the plural affix f, menD to the nounxue2shengl.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"A morpheme, on the other hand, usually corresponds to a unique hanzi, though there are a few cases where variant forms are found.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Making the ten judgements (2 types for 5 systems) takes on average 2 minutes.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For developing these mechanisms, the possibility to feed in hand-annotated information is very useful.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"This is an iterative method that improves the estimation of the parameters at each iteration.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"(See Sproat and Shih 1995.)","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Up to now, most IE researchers have been creating paraphrase knowledge (or IE patterns) by hand and for specific tasks.","label":1},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Contains digits.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"These derivation trees show how the composition operations were used to derive the final structures from elementary structures.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"As can be seen in Table 4, our training data is a lot less than those used by MENE and IdentiFinder3.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"A high-level relation is agent, which relates an animate nominal to a predicate.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The first shows how constituent features and context do not help in deciding which parser to trust.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"3 61.7 38.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Exposing this information to the decoder will lead to better hypothesis recombination.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"For all lists except locations, the lists are processed into a list of tokens (unigrams).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"To facilitate comparison with previous work, we exhaustively evaluate this grammar and two other parsing models when gold segmentation is assumed (\u00c2\u00a75).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We have described grammar state splits that significantly improve parsing performance, catalogued parsing errors, and quantified the effect of segmentation errors.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Given a PCFG grammar G and a lattice L with nodes n1 ... nk, we construct the weighted grammar GL as follows: for every arc (lexeme) l E L from node ni to node nj, we add to GL the rule [l --+ tni, tni+1, ... , tnj_1] with a probability of 1 (this indicates the lexeme l spans from node ni to node nj).","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Two annotators received training with the RST definitions and started the process with a first set of 10 texts, the results of which were intensively discussed and revised.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"(Charniak et al., 1996).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"BABAR merely identifies caseframes that frequently co-occur in coreference resolutions.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Of course, we are primarily interested in applying our techniques to languages for which no labeled resources are available.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The most accurate characterization of Chinese writing is that it is morphosyllabic (DeFrancis 1984): each hanzi represents one morpheme lexically and semantically, and one syllable phonologi\u00c2\u00ad cally.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Only tokens with initCaps not found in commonWords are tested against each list in Table 2.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"Examining the word fidanzato for the \u201cNo LP\u201d and \u201cWith LP\u201d models is particularly instructive.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We used a standard one-pass phrase-based system (Koehn et al., 2003), with the following features: relative-frequency TM probabilities in both directions; a 4-gram LM with Kneser-Ney smoothing; word-displacement distortion model; and word count.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"For example, in Northern dialects (such as Beijing), a full tone (1, 2, 3, or 4) is changed to a neutral tone (0) in the final syllable of many words: Jll donglgual 'winter melon' is often pronounced donglguaO.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The 1-bit sign is almost always negative and the 8-bit exponent is not fully used on the range of values, so in practice this corresponds to quantization ranging from 17 to 20 total bits.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"The method shares some characteristics of the decision list algorithm presented in this paper.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":".","label":0},{"source_text":"There is no global pruning.","target_text":"The following recursive equation is evaluated: Qe0 (e; S; C; j) = (2) = p(fj je) max \u00c3\u0086;e00 np(jjj0; J) p(\u00c3\u0086) p\u00c3\u0086(eje0; e00) max (S0;j0) (S0 ;Cnfjg;j0)!(S;C;j) j02Cnfjg Qe00 (e0; S0;C n fjg; j0)o: The search ends in the hypotheses (I; f1; ; Jg; j).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"2 70.7 52.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"Specifically, we assume each word type W consists of feature-value pairs (f, v).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The experimental tests are carried out on the Verbmobil task (GermanEnglish, 8000-word vocabulary), which is a limited-domain spoken-language task.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"A non-optimal analysis is shown with dotted lines in the bottom frame.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"The compact variant uses sorted arrays instead of hash tables within each node, saving some memory, but still stores full 64-bit pointers.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Within this framework, we use features intended to capture degree of generality, including the output from an SVM classifier that uses the intersection between IN and OUT as positive examples.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"It rewards matches of n-gram sequences, but measures only at most indirectly overall grammatical coherence.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"We call this pseudoprojective dependency parsing, since it is based on a notion of pseudo-projectivity (Kahane et al., 1998).","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"We believe that the underlying principles of the maximum entropy framework are suitable for exploiting information from diverse sources.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Finally, the focus\/background partition is annotated, together with the focus question that elicits the corresponding answer.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"1 53.8 47.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"The returned state s(wn1) may then be used in a followon query p(wn+1js(wn1)) that extends the previous query by one word.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"The model is often further restricted so that each source word is assigned to exactly one target word (Brown et al., 1993; Ney et al., 2000).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Note that Chang, Chen, and Chen (1991), in addition to word-frequency information, include a constraint-satisfication model, so their method is really a hybrid approach.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"There has also been work using a bootstrap- ping approach [Brin 98; Agichtein and Gravano 00; Ravichandran and Hovy 02].","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"First, we parsed the training corpus, collected all the noun phrases, and looked up each head noun in WordNet (Miller, 1990).","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our representation of contextual roles is based on information extraction patterns that are converted into simple caseframes.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"To summarize, we provided: The performance of the baseline system is similar to the best submissions in last year\u2019s shared task.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"The binary language model from Section 5.2 and text phrase table were forced into disk cache before each run.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"We see from these results that the behavior of the parametric techniques are robust in the presence of a poor parser.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Fort= 1,...,T:","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"For example, a good classifier would identify Mrs. Frank as a person, Steptoe & Johnson as a company, and Honduras as a location.","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Our results show that MENERGI performs as well as IdentiFinder when trained on comparable amount of training data.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Unlike Germann et al. (2009), we chose a model size so that all benchmarks fit comfortably in main memory.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(Berland and Charniak 99) describe a method for extracting parts of objects from wholes (e.g., &quot;speedometer&quot; from &quot;car&quot;) from a large corpus using hand-crafted patterns.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"Much confusion has been sown about Chinese writing by the use of the term ideograph, suggesting that hanzi somehow directly represent ideas.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For all lists except locations, the lists are processed into a list of tokens (unigrams).","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We incorporate instance weighting into a mixture-model framework, and find that it yields consistent improvements over a wide range of baselines.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"\u00e2\u0080\u00a2 Some tools would allow for the desired annotation mode, but are so complicated (they can be used for many other purposes as well) that annotators take a long time getting used to them.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"As can be seen in Figure 3, the phrases in the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set include completely different relationships, which are not paraphrases.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"All commentaries have been tagged with part-of-speech information using Brants\u00e2\u0080\u0099 TnT1 tagger and the Stuttgart\/Tu\u00c2\u00a8bingen Tag Set automatic analysis was responsible for this decision.)","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The same form fmnh can be segmented as f-mnh, f (\u201cthat\u201d) functioning as a reletivizer with the form mnh.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"In our grammar, features are realized as annotations to basic category labels.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"JI!","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Table 6: Incremental dev set results for the manually annotated grammar (sentences of length \u00e2\u0089\u00a4 70).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Model Overview The model starts by generating a tag assignment T for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"As we have said, parse quality decreases with sentence length.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"See Figure 3 for a screenshot of the evaluation tool.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We are currently exploring such algorithms.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"If it is made up of all capital letters, then (allCaps, zone) is set to 1.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"(b) POS tagging accuracy is lowest for maSdar verbal nouns (VBG,VN) and adjectives (e.g., JJ).","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"64 94.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"All language model queries issued by machine translation decoders follow a left-to-right pattern, starting with either the begin of sentence token or null context for mid-sentence fragments.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For MUC7, there are also no published results on systems trained on only the official training data of 200 aviation disaster articles.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The average agreement among the human judges is .76, and the average agreement between ST and the humans is .75, or about 99% of the interhuman agreement.15 One can better visualize the precision-recall similarity matrix by producing from that matrix a distance matrix, computing a classical metric multidimensional scaling (Torgerson 1958; Becker, Chambers, Wilks 1988) on that dis\u00c2\u00ad tance matrix, and plotting the first two most significant dimensions.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"This is a straightforward technique that is arguably better suited to the adaptation task than the standard method of treating representative IN sentences as queries, then pooling the match results.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The first value reports use immediately after loading while the second reports the increase during scoring. dBerkeleyLM is written in Java which requires memory be specified in advance.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"There is no relation between these two interpretations other then the fact that their surface forms coincide, and we argue that the only reason to prefer one analysis over the other is compositional.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","label":1},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The approach uses both spelling and contextual rules.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Unfortunately, we have much less data to work with than with the automatic scores.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Given a sorted array A, these other packages use binary search to find keys in O(log |A|) time.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Each word is simply tagged with the semantic classes corresponding to all of its senses.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Nodes in the trie are based on arrays sorted by vocabulary identifier.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Our results suggest that it is possible to learn accurate POS taggers for languages which do not have any annotated data, but have translations into a resource-rich language.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"For alif with hamza, normalization can be seen as another level of devocalization.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"The confidence level is then used as the belief value for the knowledge source.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Step 4.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"The model starts by generating a tag assignment for each word type in a vocabulary, assuming one tag per word.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"This PP modifies another NP, whose head is a singular noun.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"However, it is desirable if we can separate them.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"both MENE and IdentiFinder used more training data than we did (we used only the official MUC 6 and MUC7 training data).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"3.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"It was also proposed to allow annotators to skip sentences that they are unable to judge.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"We tokenize MWUs and their POS tags; this reduces the tag set size to 12.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Given parameter estimates, the label for a test example x is defined as We should note that the model in equation 9 is deficient, in that it assigns greater than zero probability to some feature combinations that are impossible.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Mikheev et al.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Purely statistical approaches have not been very popular, and so far as we are aware earlier work by Sproat and Shih (1990) is the only published instance of such an approach.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"Domain adaptation is a common concern when optimizing empirical NLP applications.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Equation 2 is an estimate of the conditional probability of the label given the feature, P(yjx).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Further, Maamouri and Bies (2004) argued that the English guidelines generalize well to other languages.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"Sorted arrays store key-value pairs in an array sorted by key, incurring no space overhead.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The best answer to this is: many research labs have very competitive systems whose performance is hard to tell apart.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Based on revision 4041, we modified Moses to print process statistics before terminating.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"However, their system is a hybrid of hand-coded rules and machine learning methods.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"Let say, if we find one system doing better on 20 of the blocks, and worse on 80 of the blocks, is it significantly worse?","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"The basic strategy is, for a given pair of entity types, to start with some examples, like several famous book title and author pairs; and find expressions which contains those names; then using the found expressions, find more author and book title pairs.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"com t 700 Mountain Avenue, 2d451, Murray Hill, NJ 07974, USA.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"(Brin 98) ,describes a system for extracting (author, book-title) pairs from the World Wide Web using an approach that bootstraps from an initial seed set of examples.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"There is a (costless) transition between the NC node and f,.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"For each language, we took the same number of sentences from the bitext as there are in its treebank, and trained a supervised feature-HMM.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"This is in general very difficult, given the extremely free manner in which Chinese given names are formed, and given that in these cases we lack even a family name to give the model confidence that it is identifying a name.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"This larger corpus was kindly provided to us by United Informatics Inc., R.O.C. a set of initial estimates of the word frequencies.9 In this re-estimation procedure only the entries in the base dictionary were used: in other words, derived words not in the base dictionary and personal and foreign names were not used.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Unexpectedly, because the ministries of treasury and education both had prepared the teacher plan together.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Automatic Paraphrase Discovery based on Context and Keywords between NE Pairs","label":1},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Future work should also extend the approach to build a complete named entity extractor - a method that pulls proper names from text and then classifies them.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"2 We now introduce a new algorithm for learning from unlabeled examples, which we will call DLCoTrain (DL stands for decision list, the term Cotrain is taken from (Blum and Mitchell 98)).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"In this particular case, all English vertices are labeled as nouns by the supervised tagger.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"These methods demonstrated the benefits of incorporating linguistic features using a log-linear parameterization, but requires elaborate machinery for training.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Not surprisingly some semantic classes are better for names than others: in our corpora, many names are picked from the GRASS class but very few from the SICKNESS class.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Word Head Of Complement POS 1 '01 inna \u00e2\u0080\u009cIndeed, truly\u00e2\u0080\u009d VP Noun VBP 2 '01 anna \u00e2\u0080\u009cThat\u00e2\u0080\u009d SBAR Noun IN 3 01 in \u00e2\u0080\u009cIf\u00e2\u0080\u009d SBAR Verb IN 4 01 an \u00e2\u0080\u009cto\u00e2\u0080\u009d SBAR Verb IN Table 1: Diacritized particles and pseudo-verbs that, after orthographic normalization, have the equivalent surface form 0 an.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"For a given partial hypothesis (C; j), the order in which the cities in C have been visited can be ignored (except j), only the score for the best path reaching j has to be stored.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Within the RST \u00e2\u0080\u009cuser community\u00e2\u0080\u009d there has also been discussion whether two levels of discourse structure should not be systematically distinguished (intentional versus informational).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Table 2 shows our complete set of results.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We report results for the best and median hyperparameter settings obtained in this way.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The method shares some characteristics of the decision list algorithm presented in this paper.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In words, the judgements are normalized, so that the average normalized judgement per judge is 3.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"We used the MUC4 terrorism corpus (MUC4 Proceedings, 1992) and news articles from the Reuter\u00e2\u0080\u0099s text collection8 that had a subject code corresponding to natural disasters.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"From the point of view of computational implementation this can be problematic, since the inclusion of non-projective structures makes the parsing problem more complex and therefore compromises efficiency and in practice also accuracy and robustness.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Also, in Information Extraction (IE), in which the system tries to extract elements of some events (e.g. date and company names of a corporate merger event), several event instances from different news articles have to be aligned even if these are expressed differently.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"The significant drop in number of pupils will begin in the fall of 2003.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"We thank United Informatics for providing us with our corpus of Chinese text, and BDC for the 'Behavior ChineseEnglish Electronic Dictionary.'","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"Each feature group can be made up of many binary features.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"It is important to bear in mind, though, that this is not an inherent limitation of the model.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Section 2.1 describes how BABAR generates training examples to use in the learning process.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The manual scores are averages over the raw unnormalized scores.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"For the seen word ir, 'gen\u00c2\u00ad erals,' there is an c:NC transduction from to the node preceding ir,; this arc has cost cost( f,) - cost(unseen(f,)), so that the cost of the whole path is the desired cost( f,).","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"When 7' is adjoined at ?I in the tree 7 we obtain a tree v&quot;.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"It is likely that the more complex cases, where path information could make a difference, are beyond the reach of the parser in most cases.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The pre terminal morphological analyses are mapped to the shortened \u00e2\u0080\u009cBies\u00e2\u0080\u009d tags provided with the tree- bank.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"In contrast to results reported in Johnson (2007), we found that the per P (Ti|T \u00e2\u0088\u0092i, \u00ce\u00b2) n (f,v)\u00e2\u0088\u0088Wi P (v|Ti, f, W \u00e2\u0088\u0092i, T \u00e2\u0088\u0092i, \u00ce\u00b2) formance of our Gibbs sampler on the basic 1TW model stabilized very quickly after about 10 full it All of the probabilities on the right-hand-side are Dirichlet, distributions which can be computed analytically given counts.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"17 They also provide a set of title-driven rules to identify names when they occur before titles such as $t. 1: xianlshengl 'Mr.' or i:l:itr!J tai2bei3 shi4zhang3 'Taipei Mayor.'","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Thus at each iteration the method induces at most n x k rules, where k is the number of possible labels (k = 3 in the experiments in this paper). step 3.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"In the third and final scheme, denoted Path, we keep the extra infor2Note that this is a baseline for the parsing experiment only (Experiment 2).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"A modified language model probability p\u00c3\u0086(eje0; e00) is defined as follows: p\u00c3\u0086(eje0; e00) =  1:0 if \u00c3\u0086 = 0 p(eje0; e00) if \u00c3\u0086 = 1 : We associate a distribution p(\u00c3\u0086) with the two cases \u00c3\u0086 = 0 and \u00c3\u0086 = 1 and set p(\u00c3\u0086 = 1) = 0:7.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"PoS tags impose a unique morphological segmentation on surface tokens and present a unique valid yield for syntactic trees.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Previous work has attempted to incorporate such constraints into token-level models via heavy-handed modifications to inference procedure and objective function (e.g., posterior regularization and ILP decoding) (Grac\u00c2\u00b8a et al., 2009; Ravi and Knight, 2009).","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"If is not initCaps, then (not-initCaps, ) is set to 1.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"The results of the manual and automatic evaluation of the participating system translations is detailed in the figures at the end of this paper.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The restriction can be expressed in terms of the number of uncovered source sentence positions to the left of the rightmost position m in the coverage set.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Because we are interested in applying our techniques to languages for which no labeled resources are available, we paid particular attention to minimize the number of free parameters and used the same hyperparameters for all language pairs.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"The original OUT counts co(s, t) are weighted by a logistic function w\u03bb(s, t): To motivate weighting joint OUT counts as in (6), we begin with the \u201cideal\u201d objective for setting multinomial phrase probabilities 0 = {p(s|t), dst}, which is the likelihood with respect to the true IN distribution pi(s, t).","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Mo del Hy per par am . E n g li s h1 1 m-1 D a n i s h1 1 m-1 D u t c h1 1 m-1 G er m a n1 1 m-1 Por tug ues e1 1 m-1 S p a ni s h1 1 m-1 S w e di s h1 1 m-1 1T W be st me dia n 45.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For queries, we uniformly sampled 10 million hits and 10 million misses.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The Bayes models were able to achieve significantly higher precision than their non-parametric counterparts.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"As a first step towards modeling transliterated names, we have collected all hanzi occurring more than once in the roughly 750 foreign names in our dictionary, and we estimate the probabil\u00c2\u00ad ity of occurrence of each hanzi in a transliteration (pTN(hanzi;)) using the maximum likelihood estimate.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"The samples from each corpus were independently evaluated.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For a description of the application of AdaBoost to various NLP problems see the paper by Abney, Schapire, and Singer in this volume.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"More complex approaches such as the relaxation technique have been applied to this problem Fan and Tsai (1988}.","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In section 5, we then evaluate the entire parsing system by training and evaluating on data from the Prague Dependency Treebank.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"2 56.2 32.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"5 \u00e2\u0080\u0098Underspecified Rhetorical Markup Language\u00e2\u0080\u0099 6 This confirms the figure given by (Schauer, Hahn.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Finally, other packages implement language model estimation while we are currently dependent on them to generate an ARPA file.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The hypothesized parse is then the set of constituents that are likely (P > 0.5) to be in the parse according to this model.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"This is the parse that is closest to the centroid of the observed parses under the similarity metric.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In the experiments below, we employ a data-driven deterministic dependency parser producing labeled projective dependency graphs,3 previously tested on Swedish (Nivre et al., 2004) and English (Nivre and Scholz, 2004).","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"To our knowledge, ours is the first analysis of this kind for Arabic parsing.","label":0},{"source_text":"Here both parametric and non-parametric models are explored.","target_text":"These two principles guide experimentation in this framework, and together with the evaluation measures help us decide which specific type of substructure to combine.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We will also directly compare with a baseline similar to the Matsoukas et al approach in order to measure the benefit from weighting phrase pairs (or ngrams) rather than full sentences.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Other approaches encode sparsity as a soft constraint.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"3 54.4 33.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"\u00e2\u0080\u0094 similar results have been observed across multiple languages.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"BABAR uses unsupervised learning to acquire this knowledge from plain text without the need for annotated training data.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Finally, we provide a realistic eval uation in which segmentation is performed both in a pipeline and jointly with parsing (\u00c2\u00a76).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"In this section, we extend state to optimize left-to-right queries.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"For \u00c3\u0086 = 0, no new target word is generated, while an additional source sentence position is covered.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Reiche\u00e2\u0080\u0099s colleagues will make sure that the concept is waterproof.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Hence we decided to select ten commentaries to form a \u00e2\u0080\u0098core corpus\u00e2\u0080\u0099, for which the entire range of annotation levels was realized, so that experiments with multi-level querying could commence.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"annotation guidelines that tell annotators what to do in case of doubt.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"3.2 The DempsterShafer Decision Model.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"In this paper, Section 2 begins by explaining how contextual role knowledge is represented and learned.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"We feel that information from a whole corpus might turn out to be noisy if the documents in the corpus are not of the same genre.","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"The taggers were trained on datasets labeled with the universal tags.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"An input ABCD can be represented as an FSA as shown in Figure 2(b).","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The CFLex and CFNet knowledge sources provide positive evidence that a candidate NP and anaphor might be coreferent.","label":0},{"source_text":"Here we present two algorithms.","target_text":"In our implementation, we make perhaps the simplest choice of weak hypothesis.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"(1997; 1999) do have some statistics that show how IdentiFinder performs when the training data is reduced.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"Third, all remaining anaphora are evaluated by 11 different knowledge sources: the four contextual role knowledge sources just described and seven general knowledge sources.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"LCFRS's share several properties possessed by the class of mildly context-sensitive formalisms discussed by Joshi (1983\/85).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"How do additional ambiguities caused by devocalization affect statistical learning?","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"Find keywords for each NE pair When we look at the contexts for each domain, we noticed that there is one or a few important words which indicate the relation between the NEs (for example, the word \u00e2\u0080\u009cunit\u00e2\u0080\u009d for the phrase \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"On the other hand, when all systems produce muddled output, but one is better, and one is worse, but not completely wrong, a judge is inclined to hand out judgements of 4, 3, and 2.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Here, all tokeninternal collocations of tags unseen in our training data are pruned away.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This left 962 examples, of which 85 were noise.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Frontier nodes are annotated by zero arty functions corresponding to elementary structures.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The CoBoost algorithm just described is for the case where there are two labels: for the named entity task there are three labels, and in general it will be useful to generalize the CoBoost algorithm to the multiclass case.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Different annotations of the same text are mapped into the same data structure, so that search queries can be formulated across annotation levels.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"In the following, we assume that this word joining has been carried out.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"We attain these results using several optimizations: hashing, custom lookup tables, bit-level packing, and state for left-to-right query patterns.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The NE tagger is a rule-based system with 140 NE categories [Sekine et al. 2004].","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Judges where excluded from assessing the quality of MT systems that were submitted by their institution.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Roughly speaking, previous work can be divided into three categories, namely purely statistical approaches, purely lexi\u00c2\u00ad cal rule-based approaches, and approaches that combine lexical information with sta\u00c2\u00ad tistical information.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"The first issue relates to the completeness of the base lexicon.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"The ratio of buckets to entries is controlled by space multiplier m > 1.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"We are therefore applying a different method, which has been used at the 2005 DARPA\/NIST evaluation.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"Since all long sentence translation are somewhat muddled, even a contrastive evaluation between systems was difficult.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Dynamic programming efficiently scores many hypotheses by exploiting the fact that an N-gram language model conditions on at most N \u2212 1 preceding words.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"Keyword detection error Even if a keyword consists of a single word, there are words which are not desirable as keywords for a domain.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"All three models evaluated in this paper incorrectly analyze the constituent as iDafa; none of the models attach the attributive adjectives properly.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"1 1 0.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"), which precludes a single universal approach to adaptation.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"(a) I f f fi * fi :1 }'l ij 1\u00c2\u00a7: {1M m m s h e n 3 m e 0 shi2 ho u4 wo 3 cai2 ne ng 2 ke4 fu 2 zh e4 ge 4 ku n4 w h a t ti m e I just be abl e ov er co m e thi s C L dif fic 'When will I be able to overcome this difficulty?'","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"For MUC6, for example, Corporate- Suffix-List is made up of ltd., associates, inc., co, corp, ltd, inc, committee, institute, commission, university, plc, airlines, co., corp. and Person-Prefix- List is made up of succeeding, mr., rep., mrs., secretary, sen., says, minister, dr., chairman, ms. . For a token that is in a consecutive sequence of init then a feature Corporate-Suffix is set to 1.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The set of candidate constituents comes from the union of all the constituents suggested by the member parsers.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Our analysis begins with a description of syntactic ambiguity in unvocalized MSA text (\u00c2\u00a72).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"We can model this probability straightforwardly enough with a probabilistic version of the grammar just given, which would assign probabilities to the individual rules.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"was done by the participants.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"Taking the intersection of languages in these resources, and selecting languages with large amounts of parallel data, yields the following set of eight Indo-European languages: Danish, Dutch, German, Greek, Italian, Portuguese, Spanish and Swedish.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Surprisingly, this effect is much less obvious for out-of-domain test data.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The bootstrap method has been critized by Riezler and Maxwell (2005) and Collins et al. (2005), as being too optimistic in deciding for statistical significant difference between systems.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"This model admits a simple Gibbs sampling algorithm where the number of latent variables is proportional to the number of word types, rather than the size of a corpus as for a standard HMM sampler (Johnson, 2007).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"When a company buys another company, a paying event can occur, but these two phrases do not indicate the same event.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"We claim that no particular morphological segmentation is a-priory more likely for surface forms before exploring the compositional nature of syntactic structures, including manifestations of various long-distance dependencies.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"\u2014 I would also like to point out to commissioner Liikanen that it is not easy to take a matter to a national court.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"The entire treebank is used in the experiment, but only primary dependencies are considered.4 In all experiments, punctuation tokens are included in the data but omitted in evaluation scores.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Furthermore we expect the label distributions on the foreign to be fairly noisy, because the graph constraints have not been taken into account yet.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"This has solutions: where pI(s|t) is derived from the IN corpus using relative-frequency estimates, and po(s|t) is an instance-weighted model derived from the OUT corpus.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The confidence intervals are computed by bootstrap resampling for BLEU, and by standard significance testing for the manual scores, as described earlier in the paper.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"However, in existing systems, this expansion come with a steep increase in model complexity.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"36.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"JI!","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"We use w erations of sampling (see Figure 2 for a depiction).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"2 61.7 64.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"Each ht is a function that predicts a label (+1 or \u20141) on examples containing a particular feature xt, while abstaining on other examples: The prediction of the strong hypothesis can then be written as We now briefly describe how to choose ht and at at each iteration.","label":0},{"source_text":"There is no global pruning.","target_text":"The computing time is given in terms of CPU time per sentence (on a 450MHz PentiumIIIPC).","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"(a) IDictionary D I D:d\/0.000 B:b\/0.000 B:b\/0.000 ( b ) ( c ) ( d ) I B e s t P a t h ( I d ( I ) o D * ) I cps:nd4.!l(l() Figure 2 An abstract example illustrating the segmentation algorithm.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"Decreasing the threshold results in higher mWER due to additional search errors.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"For example, we can easily imagine that the number of paraphrases for \u00e2\u0080\u009cA buys B\u00e2\u0080\u009d is enormous and it is not possible to create a comprehensive inventory by hand.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"2.2 Contextual Role Knowledge.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Such sequences are given additional features of A begin, A continue, or A end, and the acronym is given a feature A unique.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"When the same token is to be interpreted as a single lexeme fmnh, it may function as a single adjective \u201cfat\u201d.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"Another approach to finding paraphrases is to find phrases which take similar subjects and objects in large corpora by using mutual information of word distribution [Lin and Pantel 01].","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"08 84.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"This paper proposes a simple and effective tagging method that directly models tag sparsity and other distributional properties of valid POS tag assignments.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Proper-Name Identification.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Test set OOV rate is computed using the following splits: ATB (Chiang et al., 2006); CTB6 (Huang and Harper, 2009); Negra (Dubey and Keller, 2003); English, sections 221 (train) and section 23 (test).","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"Given this similarity function, we define a nearest neighbor graph, where the edge weight for the n most similar vertices is set to the value of the similarity function and to 0 for all other vertices.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"The probability of the classes assigned to the words in a sentence in a document is defined as follows: where is determined by the maximum entropy classifier.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"This is especially true in the case of quotations\u00e2\u0080\u0094which are common in the ATB\u00e2\u0080\u0094where (1) will follow a verb like (2) (Figure 1).","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"For the error counts, a range from 0:0 to 1:0 is used.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"The predominate focus of building systems that translate into English has ignored so far the difficult issues of generating rich morphology which may not be determined solely by local context.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We are currently working on a complete open source implementation of a training and decoding system, which should become available over the summer. pus, from which also the in-domain test set is taken.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Depending on the threshold t0, the search algorithm may miss the globally optimal path which typically results in additional translation errors.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Each xii is a member of X, where X is a set of possible features.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"In this section, we extend state to optimize left-to-right queries.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"We showed that any system defined in this way can be recognized in polynomial time.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"In this section for the purposes of showing that polynomial time recognition is possible, we make the additional restriction that the contribution of a derived structure to the input string can be specified by a bounded sequence of substrings of the input.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"All of these systems were run on data that was not seen during their development.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Finally, we incorporate the instance-weighting model into a general linear combination, and learn weights and mixing parameters simultaneously. where c\u03bb(s, t) is a modified count for pair (s, t) in OUT, u(s|t) is a prior distribution, and y is a prior weight.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":", Sun day, then the feature DayOfTheWeek is set to 1.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"More details on the parsing algorithm can be found in Nivre (2003).","label":0},{"source_text":"This topic has been getting more attention, driven by the needs of various NLP applications.","target_text":"We are focusing on phrases which have two Named Entities (NEs), as those types of phrases are very important for IE applications.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"The German finite verbs 'bin' (second example) and 'konnten' (third example) are too far away from the personal pronouns 'ich' and 'Sie' (6 respectively 5 source sentence positions).","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Many packages perform language model queries.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"The wounds are still healing.), entity-attribute (e.g., She 2001), who determined that in their corpus of German computer tests, 38% of relations were lexically signalled.","label":0},{"source_text":"There is no global pruning.","target_text":"(f1; ;mg ; l) 2 (f1; ;mg n fl; l1g ; l0) !","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"This is the only important case, because otherwise the simple majority combining technique would pick the correct constituent.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"If (wi, r, wj) E A, we say that wi is the head of wj and wj a dependent of wi.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"19 We note that it is not always clear in Wang, Li, and Chang's examples which segmented words.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"To formalize the approach, we introduce four verbgroup states S: Initial (I): A contiguous, initial block of source positions is covered.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For example, if FCC and Federal Communications Commission are both found in a document, then Federal has A begin set to 1, Communications has A continue set to 1, Commission has A end set to 1, and FCC has A unique set to 1.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"Overall, it gives improvements ranging from 1.1% for German to 14.7% for Italian, for an average improvement of 8.3% over the unsupervised feature-HMM model.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"The traveling salesman problem is an optimization problem which is defined as follows: given are a set of cities S = s1; ; sn and for each pair of cities si; sj the cost dij > 0 for traveling from city si to city sj . We are looking for the shortest tour visiting all cities exactly once while starting and ending in city s1.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The model we use provides a simple framework in which to incorporate a wide variety of lexical information in a uniform way.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The data is sorted based on the frequency of the context (\u00e2\u0080\u009ca unit of\u00e2\u0080\u009d appeared 314 times in the corpus) and the NE pair instances appearing with that context are shown with their frequency (e.g. \u00e2\u0080\u009cNBC\u00e2\u0080\u009d and \u00e2\u0080\u009cGeneral Electric Co.\u00e2\u0080\u009d appeared 10 times with the context \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"For fair comparison, we have tabulated all results with the size of training data used (Table 5 and Table 6).","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Such discrepancies can be aligned via an intermediate level of PoS tags.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"We focus on phrases which connect two Named Entities (NEs), and proceed in two stages.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"These universal POS categories not only facilitate the transfer of POS information from one language to another, but also relieve us from using controversial evaluation metrics,2 by establishing a direct correspondence between the induced hidden states in the foreign language and the observed English labels.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Most similar is scrolling queries, wherein left-to-right queries that add one word at a time are optimized.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"In this shared task, we were also confronted with this problem, and since we had no funding for paying human judgements, we asked participants in the evaluation to share the burden.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"In practice, this sparsity constraint is difficult to incorporate in a traditional POS induction system (Me\u00c2\u00b4rialdo, 1994; Johnson, 2007; Gao and Johnson, 2008; Grac\u00c2\u00b8a et al., 2009; Berg-Kirkpatrick et al., 2010).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"8 1 2.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Clearly this is not the only way to estimate word-frequencies, however, and one could consider applying other methods: in partic\u00c2\u00ad ular since the problem is similar to the problem of assigning part-of-speech tags to an untagged corpus given a lexicon and some initial estimate of the a priori probabilities for the tags, one might consider a more sophisticated approach such as that described in Kupiec (1992); one could also use methods that depend on a small hand-tagged seed corpus, as suggested by one reviewer.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"64 94.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Affix Pron Base category N found N missed (recall) N correct (precision) t,-,7 The second issue is that rare family names can be responsible for overgeneration, especially if these names are otherwise common as single-hanzi words.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"Note that in our model, conditioned on T , there is precisely one t which has nonzero probability for the token component, since for each word, exactly one \u00ce\u00b8t has support.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Another thread of relevant research has explored the use of features in unsupervised POS induction (Smith and Eisner, 2005; Berg-Kirkpatrick et al., 2010; Hasan and Ng, 2009).","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"971,746 sentences of New York Times text were parsed using the parser of (Collins 96).1 Word sequences that met the following criteria were then extracted as named entity examples: whose head is a singular noun (tagged NN).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"For example, we may have semantic information (e.g. database query operations) associated with the productions in a grammar.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"64 76.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"(2009).","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"Parameter Component As in the standard Bayesian HMM (Goldwater and Griffiths, 2007), all distributions are independently drawn from symmetric Dirichlet distributions: 2 Note that t and w denote tag and word sequences respectively, rather than individual tokens or tags.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Within the RST \u00e2\u0080\u009cuser community\u00e2\u0080\u009d there has also been discussion whether two levels of discourse structure should not be systematically distinguished (intentional versus informational).","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"One of the strengths of the DempsterShafer model is its natural ability to recognize when several credible hypotheses are still in play.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Given an anaphor and candidate, BABAR checks (1) whether the semantic classes of the anaphor intersect with the semantic expectations of the caseframe that extracts the candidate, and (2) whether the semantic classes of the candidate intersect with the semantic ex pectations of the caseframe that extracts the anaphor.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The rewrite rules and the definition of the composition operations may be stored in the finite state control since G uses a finite number of them.","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"This is less effective in our setting, where IN and OUT are disparate.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"Sometime, multiple words are needed, like \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d, \u00e2\u0080\u009cprime minister\u00e2\u0080\u009d or \u00e2\u0080\u009cpay for\u00e2\u0080\u009d (\u00e2\u0080\u009cpay\u00e2\u0080\u009d and \u00e2\u0080\u009cpay for\u00e2\u0080\u009d are different senses in the CC-domain).","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"In contrast to the Bayesian HMM, \u00ce\u00b8t is not drawn from a distribution which has support for each of the n word types.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"ICOC and CSPP contributed the greatest im provements.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Some approaches depend upon some form of con\u00c2\u00ad straint satisfaction based on syntactic or semantic features (e.g., Yeh and Lee [1991], which uses a unification-based approach).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"In Table 5 we present results from small test cor\u00c2\u00ad pora for the productive affixes handled by the current version of the system; as with names, the segmentation of morphologically derived words is generally either right or wrong.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Each lattice arc corresponds to a segment and its corresponding PoS tag, and a path through the lattice corresponds to a specific morphological segmentation of the utterance.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"6One of our experimental settings lacks document boundaries, and we used this approximation in both settings for consistency.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"The implication of this ambiguity for a parser is that the yield of syntactic trees no longer consists of spacedelimited tokens, and the expected number of leaves in the syntactic analysis in not known in advance.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"As with HG's derivation structures are annotated; in the case of TAG's, by the trees used for adjunction and addresses of nodes of the elementary tree where adjunctions occurred.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"A non-optimal analysis is shown with dotted lines in the bottom frame.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"The terms on the right-hand-side denote the type-level and token-level probability terms respectively.","label":0},{"source_text":"Their empirical results demonstrate that the type-based tagger rivals state-of-the-art tag-level taggers which employ more sophisticated learning mechanisms to exploit similar constraints.","target_text":"We can only compare with Grac\u00c2\u00b8a et al.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"In our grammar, features are realized as annotations to basic category labels.","label":0},{"source_text":"Here we present two algorithms.","target_text":"The maximum likelihood estimates (i.e., parameter values which maximize 10) can not be found analytically, but the EM algorithm can be used to hill-climb to a local maximum of the likelihood function from some initial parameter settings.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"As we have seen, the lexicon of basic words and stems is represented as a WFST; most arcs in this WFST represent mappings between hanzi and pronunciations, and are costless.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"input token, the segmentation is then performed deterministically given the 1-best analysis.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"(Hartmann 1984), for example, used the term Reliefgebung to characterize the distibution of main and minor information in texts (similar to the notion of nuclearity in RST).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"This work has been supported as part of the Verbmobil project (contract number 01 IV 601 A) by the German Federal Ministry of Education, Science, Research and Technology and as part of the Eutrans project (ESPRIT project number 30268) by the European Community.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Finally, we wish to reiterate an important point.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The manual evaluation of scoring translation on a graded scale from 1\u20135 seems to be very hard to perform.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Unfortunately, we were unable to correctly run the IRSTLM quantized variant.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"95 76.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"For effectively annotating connectives\/scopes, we found that existing annotation tools were not well-suited, for two reasons: \u00e2\u0080\u00a2 Some tools are dedicated to modes of annotation (e.g., tiers), which could only quite un-intuitively be used for connectives and scopes.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This is especially effective at reducing load time, since raw bytes are read directly to memory\u2014or, as happens with repeatedly used models, are already in the disk cache.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"structure Besides the applications just sketched, the over- arching goal of developing the PCC is to build up an empirical basis for investigating phenomena of discourse structure.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"About half of the participants of last year\u2019s shared task participated again.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Since guess and gold trees may now have different yields, the question of evaluation is complex.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"Sparse lookup is a key subproblem of language model queries.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The precision and recall measures (described in more detail in Section 3) used in evaluating Treebank parsing treat each constituent as a separate entity, a minimal unit of correctness.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Email: gale@research.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Applying the function PROJECTIVIZE to the graph in Figure 1 yields the graph in Figure 2, where the problematic arc pointing to Z has been lifted from the original head jedna to the ancestor je.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Unsupervised Models for Named Entity Classification Collins","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"This modification brings the method closer to the DL-CoTrain algorithm described earlier, and is motivated by the intuition that all three labels should be kept healthily populated in the unlabeled examples, preventing one label from dominating \u2014 this deserves more theoretical investigation.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Statistical NERs usually find the sequence of tags that maximizes the probability , where is the sequence of words in a sentence, and is the sequence of named-entity tags assigned to the words in . Attempts have been made to use global information (e.g., the same named entity occurring in different sentences of the same document), but they usually consist of incorporating an additional classifier, which tries to correct the errors in the output of a first NER (Mikheev et al., 1998; Borthwick, 1999).","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"The performance was 80.99% recall and 61.83% precision.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"The annotator can then \u00e2\u0080\u009cclick away\u00e2\u0080\u009d those words that are here not used as connectives (such as the conjunction und (\u00e2\u0080\u0098and\u00e2\u0080\u0099) used in lists, or many adverbials that are ambiguous between connective and discourse particle).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"Preferences for constituent order (especially in languages with relatively free word order) often belong to this group.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"The family name set is restricted: there are a few hundred single-hanzi family names, and about ten double-hanzi ones.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"The importance of dictionaries in NERs has been investigated in the literature (Mikheev et al., 1999).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"I \u00e2\u0080\u00a2 JAPANS :rl4 .\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\)\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"o\u00c2\u00b7'\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\"\\:J\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7 \u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7\u00c2\u00b7'\\; . '.:: ..........0 6.51 9.51 : jj \/ JAPANESE OCTOPUS 10\u00c2\u00b728i\u00c2\u00a3 :_nc HOW SAY f B :rl4 :il: :wen2 t '- \u00e2\u0080\u00a2 :zhang!","label":0},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"The baseline system in Table 3 refers to the maximum entropy system that uses only local features.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"Consider the following sentences: (a) Jose Maria Martinez, Roberto Lisandy, and Dino Rossy, who were staying at a Tecun Uman hotel, were kidnapped by armed men who took them to an unknown place.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"95 76.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Only tokens with initCaps not found in commonWords are tested against each list in Table 2.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We now describe the CoBoost algorithm for the named entity problem.","label":0},{"source_text":"BABAR's performance in both domains of terrorism and natural disaster, and the contextual-role knowledge in pronouns have shown successful results.","target_text":"The CFLex and CFNet knowledge sources provide positive evidence that a candidate NP and anaphor might be coreferent.","label":0},{"source_text":"This paper talks about KenLM: Faster and Smaller Language Model Queries.","target_text":"These performance gains transfer to improved system runtime performance; though we focused on Moses, our code is the best lossless option with cdec and Joshua.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"The algorithm, called CoBoost, has the advantage of being more general than the decision-list learning alInput: (xi , yi), , (xim, ) ; x, E 2x, yi = +1 Initialize Di (i) = 1\/m.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"Two aspects of the corpus have been presented in previous papers ((Re- itter, Stede 2003) on underspecified rhetorical structure; (Stede 2003) on the perspective of knowledge-based summarization).","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"Also, expanding on the techniques for the automatic generation of extraction patterns (Riloff 96; Sudo 03) using our method, the extraction patterns which have the same meaning can be automatically linked, enabling us to produce the final table fully automatically.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Historically, Arabic grammar has identified two sentences types: those that begin with a nominal (\u00ef\u00bf\u00bd '.i \u00ef\u00bf\u00bdu _..","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"While the linear precedence of segmental morphemes within a token is subject to constraints, the dominance relations among their mother and sister constituents is rather free.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"2.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"Precision is the portion of hypothesized constituents that are correct and recall is the portion of the Treebank constituents that are hypothesized.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"example, in Northern Mandarin dialects there is a morpheme -r that attaches mostly to nouns, and which is phonologically incorporated into the syllable to which it attaches: thus men2+r (door+R) 'door' is realized as mer2.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Alon Lavie advised on this work.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"We define the following function: If Zco is small, then it follows that the two classifiers must have a low error rate on the labeled examples, and that they also must give the same label on a large number of unlabeled instances.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seemed to be less pronounced on out-of-domain test data.","target_text":"This work was supported in part under the GALE program of the Defense Advanced Research Projects Agency, Contract No.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"Je voudrais pr\u00b4eciser, a` l\u2019adresse du commissaire Liikanen, qu\u2019il n\u2019est pas ais\u00b4e de recourir aux tribunaux nationaux.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"As a result, Arabic sentences are usually long relative to English, especially after Length English (WSJ) Arabic (ATB) \u00e2\u0089\u00a4 20 41.9% 33.7% \u00e2\u0089\u00a4 40 92.4% 73.2% \u00e2\u0089\u00a4 63 99.7% 92.6% \u00e2\u0089\u00a4 70 99.9% 94.9% Table 2: Frequency distribution for sentence lengths in the WSJ (sections 2\u00e2\u0080\u009323) and the ATB (p1\u00e2\u0080\u00933).","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In general, m; l; l0 6= fl1; l2; l3g and in line umber 3 and 4, l0 must be chosen not to violate the above reordering restriction.","label":0},{"source_text":"Their method did not assume any knowledge about the target language, making it applicable to a wide array of resource-poor languages.","target_text":"We use label propagation in two stages to generate soft labels on all the vertices in the graph.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"For example, even if the contexts surrounding an anaphor and candidate match exactly, they are not coreferent if they have substantially different meanings 9 We would be happy to make our manually annotated test data available to others who also want to evaluate their coreference resolver on the MUC4 or Reuters collections.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"For example, in Information Retrieval (IR), we have to match a user\u00e2\u0080\u0099s query to the expressions in the desired documents, while in Question Answering (QA), we have to find the answer to the user\u00e2\u0080\u0099s question even if the formulation of the answer in the document is different from the question.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Our results outperform strong unsupervised baselines as well as approaches that rely on direct projections, and bridge the gap between purely supervised and unsupervised POS tagging models.","label":1},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"(2010) reports the best unsupervised results for English.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"Keywords with more than one word In the evaluation, we explained that \u00e2\u0080\u009cchairman\u00e2\u0080\u009d and \u00e2\u0080\u009cvice chairman\u00e2\u0080\u009d are considered paraphrases.","label":0},{"source_text":"In this paper, the authors are of the opinion that the sequence models-based approaches usually treat token-level tag assignment as the primary latent variable.","target_text":"For instance, on Spanish, the absolute gap on median performance is 10%.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"Call the crossing constituents A and B.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"It would have therefore also been possible to use the integer programming (IP) based approach of Ravi and Knight (2009) instead of the feature-HMM for POS induction on the foreign side.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Indeed there are several open issues.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"The pseudo-code describing the algorithm is given in Fig.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"More formally, we start by representing the dictionary D as a Weighted Finite State Trans\u00c2\u00ad ducer (WFST) (Pereira, Riley, and Sproat 1994).","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"Next we show that the ATB is similar to other tree- banks in gross statistical terms, but that annotation consistency remains low relative to English (\u00c2\u00a73).","label":1},{"source_text":"NER is useful in many NLP applications such as information extraction, question answering, etc .Chieu and Ng have shown that the maximum entropy framework is able to use global information directly from various sources.","target_text":"Statistical NERs usually find the sequence of tags that maximizes the probability , where is the sequence of words in a sentence, and is the sequence of named-entity tags assigned to the words in . Attempts have been made to use global information (e.g., the same named entity occurring in different sentences of the same document), but they usually consist of incorporating an additional classifier, which tries to correct the errors in the output of a first NER (Mikheev et al., 1998; Borthwick, 1999).","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Furthermore, even the size of the dictionary per se is less important than the appropriateness of the lexicon to a particular test corpus: as Fung and Wu (1994) have shown, one can obtain substantially better segmentation by tailoring the lexicon to the corpus to be segmented.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"attaching to terms denoting human beings.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Finally, a DempsterShafer probabilistic model evaluates the evidence provided by the knowledge sources for all candidate antecedents and makes the final resolution decision.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Then, token- level HMM emission parameters are drawn conditioned on these assignments such that each word is only allowed probability mass on a single assigned tag.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Nodes in the trie are based on arrays sorted by vocabulary identifier.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"Three state-of-the-art statistical parsers are combined to produce more accurate parses, as well as new bounds on achievable Treebank parsing accuracy.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"example, in Northern Mandarin dialects there is a morpheme -r that attaches mostly to nouns, and which is phonologically incorporated into the syllable to which it attaches: thus men2+r (door+R) 'door' is realized as mer2.","label":0},{"source_text":"Finally, several coreference systems have successfully incorporated anaphoricity determination modules.","target_text":"3.2 The DempsterShafer Decision Model.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Thus an explicit assumption about the redundancy of the features \u2014 that either the spelling or context alone should be sufficient to build a classifier \u2014 has been built into the algorithm.","label":0},{"source_text":"This paper discusses the Potsdam Commentary Corpus, a corpus of german assembeled by potsdam university.","target_text":"Unexpectedly, because the ministries of treasury and education both had prepared the teacher plan together.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"When dropping the top and bottom 2.5% the remaining BLEU scores define the range of the confidence interval.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"For that application, at a minimum, one would want to know the phonological word boundaries.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Note that the good classes JADE, GOLD and GRASS have lower costs than the bad classes SICKNESS, DEATH and RAT, as desired, so the trend observed for the results of this method is in the right direction.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"However, when the linear LM is combined with a linear TM (lm+lin tm) or MAP TM (lm+map TM), the results are much better than a log-linear combination for the EMEA setting, and on a par for NIST.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We experiment with four values for each hyperparameter resulting in 16 (\u00ce\u00b1, \u00ce\u00b2) combinations: \u00ce\u00b1 \u00ce\u00b2 0.001, 0.01, 0.1, 1.0 0.01, 0.1, 1.0, 10 Iterations In each run, we performed 30 iterations of Gibbs sampling for the type assignment variables W .4 We use the final sample for evaluation.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"(2009) on Portuguese (Grac\u00c2\u00b8a et al.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"We show translation results for three approaches: the monotone search (MonS), where no word reordering is allowed (Tillmann, 1997), the quasimonotone search (QmS) as presented in this paper and the IBM style (IbmS) search as described in Section 3.2.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Since guess and gold trees may now have different yields, the question of evaluation is complex.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"In this domain the major scenarios involve the things they agreed on, rather than the mere fact that they agreed.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"\u00e2\u0080\u009d, because Bush is the first word, the initial caps might be due to its position (as in \u00e2\u0080\u009cThey put a freeze on . . .","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"This is because different judges focused on different language pairs.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"Section 5 covers relevant previous work on SMT adaptation, and section 6 concludes.","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Models that employ this strategy are denoted hsp.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"When extracting the vector t, used to compute the constraint feature from the graph, we tried three threshold values for r (see Eq.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"This departure from the traditional token-based tagging approach allows us to explicitly capture type- level distributional properties of valid POS tag as signments as part of the model.","label":1},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"Segmental morphology Hebrew consists of seven particles m(\u201cfrom\u201d) f(\u201cwhen\u201d\/\u201cwho\u201d\/\u201cthat\u201d) h(\u201cthe\u201d) w(\u201cand\u201d) k(\u201clike\u201d) l(\u201cto\u201d) and b(\u201cin\u201d). which may never appear in isolation and must always attach as prefixes to the following open-class category item we refer to as stem.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"1 is given in Fig.","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"It is unique, agrees with the maximum-likelihood distribution, and has the exponential form (Della Pietra et al., 1997): where refers to the outcome, the history (or context), and is a normalization function.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"30 16.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"8 66.4 52.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Sequence of Initial Caps (SOIC): In the sentence Even News Broadcasting Corp., noted for its accurate reporting, made the erroneous announcement., a NER may mistake Even News Broadcasting Corp. as an organization name.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"We incur some additional memory cost due to storing state in each hypothesis, though this is minimal compared with the size of the model itself.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"This has led previous workers to adopt ad hoc linear weighting schemes (Finch and Sumita, 2008; Foster and Kuhn, 2007; L\u00a8u et al., 2007).","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Table 1 The cost as a novel given name (second position) for hanzi from various radical classes.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"This revealed interesting clues about the properties of automatic and manual scoring.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Unique Occurrences and Zone (UNIQ): This group of features indicates whether the word is unique in the whole document.","label":0},{"source_text":"The fact that projective dependency parsers can never exactly reproduce the analyses found in non-projective tree banks is often neglected because of the relative scarcity of problematic constructions.","target_text":"As expected, the most informative encoding, Head+Path, gives the highest accuracy with over 99% of all non-projective arcs being recovered correctly in both data sets.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"To conserve memory at the expense of accuracy, values may be quantized using q bits per probability and r bits per backoff6.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Consider the following sentences: (a) Jose Maria Martinez, Roberto Lisandy, and Dino Rossy, who were staying at a Tecun Uman hotel, were kidnapped by armed men who took them to an unknown place.","label":0},{"source_text":"They used graph-based label propagation for cross-lingual knowledge transfer and used the projected labels as features in an unsupervised model.","target_text":"We would like to thank Ryan McDonald for numerous discussions on this topic.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"gaolbu4-gaolxing4 (hap-not-happy) 'happy?'","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"\u00e2\u0080\u00a2 Bridging links: the annotator is asked to specify the type as part-whole, cause-effect (e.g., She had an accident.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"3 68.9 50.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The method uses a &quot;soft&quot; measure of the agreement between two classifiers as an objective function; we described an algorithm which directly optimizes this function.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"We reduce this to O(log log |A|) time by evenly distributing keys over their range then using interpolation search4 (Perl et al., 1978).","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The normalized judgement per judge is the raw judgement plus (3 minus average raw judgement for this judge).","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"The semantic agreement KS eliminates some candidates, but also provides positive evidence in one case: if the candidate and anaphor both have semantic tags human, company, date, or location that were assigned via NER or the manually labeled dictionary entries.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"7 Big 5 is the most popular Chinese character coding standard in use in Taiwan and Hong Kong.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Lexicon Feature of Previous and Next Token: The string of the previous token and the next token is used with the initCaps information of . If has initCaps, then a feature (initCaps, ) is set to 1.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Compared to last year\u2019s shared task, the participants represent more long-term research efforts.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"3) A tight coupling with the speech recognizer output.","label":0},{"source_text":"Their work is closely related to recent approaches that incorporate the sparsity constraint into the POS induction process.","target_text":"The hyperparameters \u00ce\u00b1 and \u00ce\u00b2 represent the concentration parameters of the token- and type-level components of the model respectively.","label":0},{"source_text":"Their results suggested that it was possible to learn accurate POS taggers for languages which did not have any annotated data, but have translations into a resource-rich language.","target_text":"This locally normalized log-linear model can look at various aspects of the observation x, incorporating overlapping features of the observation.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"We refer to (T , W ) as the lexicon of a language and \u00cf\u0088 for the parameters for their generation; \u00cf\u0088 depends on a single hyperparameter \u00ce\u00b2.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"Case and Zone: If the token starts with a capital letter (initCaps), then an additional feature (init- Caps, zone) is set to 1.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Although the segmentation requirements for Arabic are not as extreme as those for Chinese, Arabic is written with certain cliticized prepositions, pronouns, and connectives connected to adjacent words.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"For even larger models, we recommend RandLM; the memory consumption of the cache is not expected to grow with model size, and it has been reported to scale well.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"However, it is almost universally the case that no clear definition of what constitutes a \"correct\" segmentation is given, so these performance measures are hard to evaluate.","label":0},{"source_text":"The corpus was annoted with different linguitic information.","target_text":"2.1 Part-of-speech tags.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"An anti-greedy algorithm, AG: instead of the longest match, take the.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"We can do that . IbmS: Yes, wonderful.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The tree t2 must be on one of the two branches.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Variants of alif are inconsistently used in Arabic texts.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"Pairwise comparison: We can use the same method to assess the statistical significance of one system outperforming another.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"This range is collapsed to a number of buckets, typically by taking the hash modulo the number of buckets.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"First, we learn weights on individual phrase pairs rather than sentences.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"A contextual rule considers words surrounding the string in the sentence in which it appears (e.g., a rule that any proper name modified by an appositive whose head is president is a person).","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"The entries in such a lexicon may be thought of as meaningful surface segments paired up with their PoS tags li = (si, pi), but note that a surface segment s need not be a space-delimited token.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"As an indication, in our core corpus, we found an average sentence length of 15.8 words and 1.8 verbs per sentence, whereas a randomly taken sample of ten commentaries from the national papers Su\u00c2\u00a8ddeutsche Zeitung and Frankfurter Allgemeine has 19.6 words and 2.1 verbs per sentence.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"The wounds are still healing.), entity-attribute (e.g., She 2001), who determined that in their corpus of German computer tests, 38% of relations were lexically signalled.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"With regard to purely morphological phenomena, certain processes are not han\u00c2\u00ad dled elegantly within the current framework Any process involving reduplication, for instance, does not lend itself to modeling by finite-state techniques, since there is no way that finite-state networks can directly implement the copying operations required.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"Note that in our formalism a weakhypothesis can abstain.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"For some language pairs (such as GermanEnglish) system performance is more divergent than for others (such as English-French), at least as measured by BLEU.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"With the exception of the Dutch data set, no other processing is performed on the annotated tags.","label":0},{"source_text":"Foster et all describe a new approach to SMT adaptation that weights out-of-domain phrase pairs according to their relevance to the target domain, determined by both how similar to it they appear to be, and whether they belong to general language or not.","target_text":"We have already mentioned the closely related work by Matsoukas et al (2009) on discriminative corpus weighting, and Jiang and Zhai (2007) on (nondiscriminative) instance weighting.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"Replacing this with an ranked evaluation seems to be more suitable.","label":1},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Machine Translation In this case my colleague can not visit on I n d i e s e m F a l l ka nn m e i n K o l l e g e a m the v i e r t e n M a i n i c h t b e s u c h e n S i e you fourth of May Figure 1: Reordering for the German verbgroup.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"The Wang, Li, and Chang system fails on fragment (b) because their system lacks the word youlyoul 'soberly' and misinterpreted the thus isolated first youl as being the final hanzi of the preceding name; similarly our system failed in fragment (h) since it is missing the abbreviation i:lJI!","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Nonstochastic lexical-knowledge-based approaches have been much more numer\u00c2\u00ad ous.","label":0},{"source_text":"Explanations for this phenomenon are relative informativeness of lexicalization, insensitivity to morphology and the effect of variable word order and these factors lead to syntactic disambiguation.","target_text":"Cohen and Smith (2007) chose a metric like SParseval (Roark et al., 2006) that first aligns the trees and then penalizes segmentation errors with an edit-distance metric.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"In general, the lemma of the previous section does not ensure that all the productions in the combined parse are found in the grammars of the member parsers.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"The learning task is to find two classifiers : 2x1 { \u20141, +1} 12 : 2x2 { \u20141, +1} such that (x1,) = f2(x2,t) = yt for examples i = 1, , m, and f1 (x1,) = f2 (x2,t) as often as possible on examples i = m + 1, ,n. To achieve this goal we extend the auxiliary function that bounds the training error (see Equ.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"This is in contrast to dependency treebanks, e.g.","label":0},{"source_text":"The second algorithm builds on a boosting algorithm called AdaBoost.","target_text":"The algorithm, called CoBoost, has the advantage of being more general than the decision-list learning alInput: (xi , yi), , (xim, ) ; x, E 2x, yi = +1 Initialize Di (i) = 1\/m.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"Gazdar (1985) argues that sharing of stacks can be used to give analyses for coordination.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The experimental tests are carried out on the Verbmobil task (GermanEnglish, 8000-word vocabulary), which is a limited-domain spoken-language task.","label":1},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"We dropped, however, one of the languages, Finnish, partly to keep the number of tracks manageable, partly because we assumed that it would be hard to find enough Finnish speakers for the manual evaluation.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Confidence Interval: To estimate confidence intervals for the average mean scores for the systems, we use standard significance testing.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Now, for this application one might be tempted to simply bypass the segmentation problem and pronounce the text character-by-character.","label":0},{"source_text":"They showed that it was useful to abstract away from the details of the formalism, and examine the nature of their derivation process as reflected by properties their trees, find that several of the formalisms considered can be seen as being closely related since they have derivation tree sets with the same structure as those produced by Context-Free Grammars.","target_text":"The space of a configuration is the sum of the lengths of the nonblank tape contents of the k work tapes.","label":0},{"source_text":"They employed a PCFG-based generative framework to make both syntactic and morphological disambiguation decisions which were not only theoretically clean and linguistically justified but also probabilistically appropriate and empirically sound.","target_text":"There is no relation between these two interpretations other then the fact that their surface forms coincide, and we argue that the only reason to prefer one analysis over the other is compositional.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"We are interested in combining the substructures of the input parses to produce a better parse.","label":0},{"source_text":"The features were weighted within a logistic model that gave an overall weight that was applied to the phrase pair and MAP-smoothed relative-frequency estimates which were combined linearly with relative-frequency estimates from an in-domain phrase table.","target_text":"6One of our experimental settings lacks document boundaries, and we used this approximation in both settings for consistency.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(See also Wu and Fung [1994].)","label":0},{"source_text":"Using a treebank grammar, the data-driven lexicon, and a linguistically motivated unknown-tokens handling technique their model outperformed previous pipelined, integrated or factorized systems for Hebrew morphological and syntactic processing, yielding an error reduction of 12% over the best published results so far.","target_text":"The additional morphological material in such cases appears after the stem and realizes the extended meaning.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"A second restriction on the formalisms is that choices during the derivation are independent of the context in the derivation.","label":0},{"source_text":"Using less training data than other systems, their NER can perform as well as other state-of-the-art NERs.","target_text":"MENE is then trained on 80% of the training corpus, and tested on the remaining 20%.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Each lattice arc corresponds to a segment and its corresponding PoS tag, and a path through the lattice corresponds to a specific morphological segmentation of the utterance.","label":0},{"source_text":"We chose one of four labels for each example: location, person, organization, or noise where the noise category was used for items that were outside the three categories.","target_text":"For the experiments in this paper we made a couple of additional modifications to the CoBoost algorithm.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The learning task is to find two classifiers : 2x1 { \u20141, +1} 12 : 2x2 { \u20141, +1} such that (x1,) = f2(x2,t) = yt for examples i = 1, , m, and f1 (x1,) = f2 (x2,t) as often as possible on examples i = m + 1, ,n. To achieve this goal we extend the auxiliary function that bounds the training error (see Equ.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"In contrast to these approaches, our method directly incorporates these constraints into the structure of the model.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"of Tokens No.","label":0},{"source_text":"There is no global pruning.","target_text":"The cities of the traveling salesman problem correspond to source Table 1: DP algorithm for statistical machine translation.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"The average individual parser accuracy was reduced by more than 5% when we added this new parser, but the precision of the constituent voting technique was the only result that decreased significantly.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"including Third Tone Sandhi (Shih 1986), which changes a 3 (low) tone into a 2 (rising) tone before another 3 tone: 'j\";gil, xiao3 [lao3 shu3] 'little rat,' becomes xiao3 { lao2shu3 ], rather than xiao2 { lao2shu3 ], because the rule first applies within the word lao3shu3 'rat,' blocking its phrasal application.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Since these are distinct syntactic units, they are typically segmented.","label":0},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"When the connective is an adverbial, there is much less clarity as to the range of the spans.","label":0},{"source_text":"The three parsers were trained and tuned by their creators on various sections of the WSJ portion of the Penn Treebank.","target_text":"When this metric is less than 0.5, we expect to incur more errors' than we will remove by adding those constituents to the parse.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"8 We use head-finding rules specified by a native speaker.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The way judgements are collected, human judges tend to use the scores to rank systems against each other.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"Each visited entry wni stores backoff b(wni ).","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":".., Tn ) T W \u00cf\u0086 E w : Token word seqs (obs) t : Token tag assigns (det by T ) PARAMETERS \u00cf\u0088 : Lexicon parameters \u00ce\u00b8 : Token word emission parameters \u00cf\u0086 : Token tag transition parameters \u00cf\u0086 \u00cf\u0086 t1 t2 \u00ce\u00b8 \u00ce\u00b8 w1 w2 K \u00cf\u0086 T tm O K \u00ce\u00b8 E wN m N N Figure 1: Graphical depiction of our model and summary of latent variables and parameters.","label":0},{"source_text":"This paper conducted research in the area of automatic paraphrase discovery.","target_text":"The sentences in the corpus were tagged by a transformation-based chunker and an NE tagger.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"Here we present two algorithms.","target_text":"(3)) to be defined over unlabeled as well as labeled instances.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"This best instance-weighting model beats the equivalant model without instance weights by between 0.6 BLEU and 1.8 BLEU, and beats the log-linear baseline by a large margin.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"123 examples fell into the noise category.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"The complexity of the algorithm is O(E3 J2 2J), where E is the size of the target language vocabulary.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"There may be occasionally a system clearly at the top or at the bottom, but most systems are so close that it is hard to distinguish them.","label":0},{"source_text":"This paper talks about Unsupervised Models for Named Entity Classification.","target_text":"Fig.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Proper-Name Identification.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The scores and confidence intervals are detailed first in the Figures 7\u201310 in table form (including ranks), and then in graphical form in Figures 11\u201316.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"Our work is motivated by the observation that contextual roles can be critically important in determining the referent of a noun phrase.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"They first collect the NE instance pairs and contexts, just like our method.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Our code has been publicly available and intergrated into Moses since October 2010.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Such discrepancies can be aligned via an intermediate level of PoS tags.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"The general idea for the knowledge- based part is to have the system use as much information as it can find at its disposal to produce a target representation as specific as possible and as underspecified as necessary.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In total, for the 2,000 NE category pairs, 5,184 keywords are found.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"We used the MUC4 terrorism corpus (MUC4 Proceedings, 1992) and news articles from the Reuter\u00e2\u0080\u0099s text collection8 that had a subject code corresponding to natural disasters.","label":0},{"source_text":"Combining multiple highly-accurate independent parsers yields promising results.","target_text":"It is closer to the smaller value of precision and recall when there is a large skew in their values.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"All experiments use ATB parts 1\u00e2\u0080\u00933 divided according to the canonical split suggested by Chiang et al.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"(a) I f f fi * fi :1 }'l ij 1\u00c2\u00a7: {1M m m s h e n 3 m e 0 shi2 ho u4 wo 3 cai2 ne ng 2 ke4 fu 2 zh e4 ge 4 ku n4 w h a t ti m e I just be abl e ov er co m e thi s C L dif fic 'When will I be able to overcome this difficulty?'","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"On MUC6 data, MENERGI also achieves performance comparable to IdentiFinder when trained on similar amount of training data.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"The simplest version of the maximum matching algorithm effectively deals with ambiguity by ignoring it, since the method is guaranteed to produce only one segmentation.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Given an anaphor, BABAR identifies the caseframe that would extract it from its sentence.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"Smith estimates Lotus will make a profit this quarter\u00e2\u0080\u00a6\u00e2\u0080\u009d, our system extracts \u00e2\u0080\u009cSmith esti mates Lotus\u00e2\u0080\u009d as an instance.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"The accuracies for link were 73% and 86% on two evaluated domains.","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"In total, across all domains, we kept 13,976 phrases with keywords.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"While we had up to 11 submissions for a translation direction, we did decide against presenting all 11 system outputs to the human judge.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"Our full model outperforms the \u201cNo LP\u201d setting because it has better vocabulary coverage and allows the extraction of a larger set of constraint features.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"The algorithm builds two classifiers iteratively: each iteration involves minimization of a continuously differential function which bounds the number of examples on which the two classifiers disagree.","label":0},{"source_text":"They proposed an unsupervised method to discover paraphrases from a large untagged corpus.","target_text":"In this section, we will explain the algorithm step by step with examples.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"This approach is similar to BABAR in that they both acquire knowledge from earlier resolutions.","label":0},{"source_text":"They found replacing it with a ranked evaluation to be more suitable.","target_text":"In all figures, we present the per-sentence normalized judgements.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In order to handle the necessary word reordering as an optimization problem within our dynamic programming approach, we describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming (Held, Karp, 1962).","label":1},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"9 www.ling.unipotsdam.de\/sfb\/ Figure 2: Screenshot of Annis Linguistic Database 3.3 Symbolic and knowledge-based.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"This kind of dependency arises from the use of the composition operation to compose two arbitrarily large categories.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"The algorithm works due to the fact that not all permutations of cities have to be considered explicitly.","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Almost all annotators reported difficulties in maintaining a consistent standard for fluency and adequacy judgements, but nevertheless most did not explicitly move towards a ranking-based evaluation.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"Experiments are presented in section 4.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"5 68.1 34.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Examples are given in Table 4.","label":0},{"source_text":"They showed better grammars to improve performance on both morphological and syntactic tasks, providing support for the advantage of a joint framework over pipelined or factorized ones.","target_text":"Such resources exist for Hebrew (Itai et al., 2006), but unfortunately use a tagging scheme which is incompatible with the one of the Hebrew Treebank.s For this reason, we use a data-driven morphological analyzer derived from the training data similar to (Cohen and Smith, 2007).","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"HR0011-06-C-0022.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"The problem with these styles of evaluation is that, as we shall demonstrate, even human judges do not agree perfectly on how to segment a given text.","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Since 170 annotated texts constitute a fairly small training set, Reitter found that an overall recognition accuracy of 39% could be achieved using his method.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"Instead, we condition on the type-level tag assignments T . Specifically, let St = {i|Ti = t} denote the indices of theword types which have been assigned tag t accord ing to the tag assignments T . Then \u00ce\u00b8t is drawn from DIRICHLET(\u00ce\u00b1, St), a symmetric Dirichlet which only places mass on word types indicated by St. This ensures that each word will only be assigned a single tag at inference time (see Section 4).","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We use the universal POS tagset of Petrov et al. (2011) in our experiments.10 This set C consists of the following 12 coarse-grained tags: NOUN (nouns), VERB (verbs), ADJ (adjectives), ADV (adverbs), PRON (pronouns), DET (determiners), ADP (prepositions or postpositions), NUM (numerals), CONJ (conjunctions), PRT (particles), PUNC (punctuation marks) and X (a catch-all for other categories such as abbreviations or foreign words).","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"In the experiments below, we employ a data-driven deterministic dependency parser producing labeled projective dependency graphs,3 previously tested on Swedish (Nivre et al., 2004) and English (Nivre and Scholz, 2004).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"5 Related Work.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Confidence Interval: Since BLEU scores are not computed on the sentence level, traditional methods to compute statistical significance and confidence intervals do not apply.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"REL+VB) (cf.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"In Figure 4, we displayed the number of system comparisons, for which we concluded statistical significance.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Another way to view the judgements is that they are less quality judgements of machine translation systems per se, but rankings of machine translation systems.","label":0},{"source_text":"Due to many similarly performing systems, they are not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The easiest language pair according to BLEU (English-French: 28.33) received worse manual scores than the hardest (English-German: 14.01).","label":0},{"source_text":"The texts were annotated with the RSTtool.","target_text":"Preferences for constituent order (especially in languages with relatively free word order) often belong to this group.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Starting from a DP-based solution to the traveling salesman problem, we present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","label":1},{"source_text":"the \"Potsdam Commentary Corpus\" or PCC consists of 170 commentaries from Ma\u00a8rkische Allgemeine Zeitung, a German regional daily.","target_text":"A different but supplementary perspective on discourse-based information structure is taken 11ventionalized patterns (e.g., order of informa by one of our partner projects, which is inter tion in news reports).","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The domain is general politics, economics and science.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Entries landing in the same bucket are said to collide.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"This is not ideal for some applications, however.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"\u2014 I would also like to point out to commissioner Liikanen that it is not easy to take a matter to a national court.","label":0},{"source_text":"A beam search concept is applied as in speech recognition.","target_text":"When translating the sentence monotonically from left to right, the translation of the German finite verb 'kann', which is the left verbal brace in this case, is postponed until the German noun phrase 'mein Kollege' is translated, which is the subject of the sentence.","label":0},{"source_text":"Due to many similarly performing systems, the author was not able to draw strong conclusions on the question of correlation of manual and automatic evaluation metrics.","target_text":"The normalization on a per-judge basis gave very similar ranking, only slightly less consistent with the ranking from the pairwise comparisons.","label":0},{"source_text":"The main result is that the combined system can recover non-projective dependencies with a precision sufficient to give a significant improvement in overall parsing accuracy, especially with respect to the exact match criterion, leading to the best reported performance for robust non-projective parsing of Czech.","target_text":"Special thanks to Jan Haji\u02c7c and Matthias Trautner Kromann for assistance with the Czech and Danish data, respectively, and to Jan Haji\u02c7c, Tom\u00b4a\u02c7s Holan, Dan Zeman and three anonymous reviewers for valuable comments on a preliminary version of the paper.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"The importance of this property becomes clear in contrasting theories underlying GPSG (Gazdar, Klein, Pulluna, and Sag, 1985), and GB (as described by Berwick, 1984) with those underlying LFG and FUG.","label":0},{"source_text":"Here we present two algorithms.","target_text":"We can now add a new weak hypothesis 14 based on a feature in X1 with a confidence value al hl and atl are chosen to minimize the function We now define, for 1 <i <n, the following virtual distribution, As before, Ztl is a normalization constant.","label":0},{"source_text":"they showed that the average agreement among the human judges is .76, and the average agreement between ST(system) and the humans is .75, or about 99% of the interhuman agreement.","target_text":"52 15.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"5.1 Data Sets.","label":0},{"source_text":"In this paper, Ben and Riloff present a coreference resolver called BABAR that focuses on the use of contextual-role knowledge for coreference resolution.","target_text":"Section 3 describes the complete coreference resolution model, which uses the contextual role knowledge as well as more traditional coreference features.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"input token, the segmentation is then performed deterministically given the 1-best analysis.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"We evaluated the results based on two metrics.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"However, ince we extracted the test corpus automatically from web sources, the reference translation was not always accurate \u2014 due to sentence alignment errors, or because translators did not adhere to a strict sentence-by-sentence translation (say, using pronouns when referring to entities mentioned in the previous sentence).","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The sentences in the corpus were tagged by a transformation-based chunker and an NE tagger.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"This is similar to stacking the different feature instantiations into long (sparse) vectors and computing the cosine similarity between them.","label":0},{"source_text":"This paper presents methods to query N-gram language models, minimizing time and space costs.","target_text":"It also does not prune, so comparing to our pruned model would be unfair.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Of course, we.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"f, nan2gual+men0 'pumpkins' is by no means impossible.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Previous work that involves the gathering of information from the whole document often uses a secondary classifier, which corrects the mistakes of a primary sentence- based classifier.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The counts represent portions of the approximately 44000 constituents hypothesized by the parsers in the development set.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For the automatic scoring method BLEU, we can distinguish three quarters of the systems.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"This feature ft incorporates information from the smoothed graph and prunes hidden states that are inconsistent with the thresholded vector tx.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"The learned patterns are then normalized and applied to the corpus.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"For that application, at a minimum, one would want to know the phonological word boundaries.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"(2010) reports the best unsupervised results for English.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"Sentences and systems were randomly selected and randomly shuffled for presentation.","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"All features were conjoined with the state z.","label":0},{"source_text":"They incorporated instance-weighting into a mixture-model framework, and found that it yielded consistent improvements over a wide range of baselines.","target_text":"Finally, we intend to explore more sophisticated instanceweighting features for capturing the degree of generality of phrase pairs.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"Unfortunately, Yarowsky's method is not well understood from a theoretical viewpoint: we would like to formalize the notion of redundancy in unlabeled data, and set up the learning task as optimization of some appropriate objective function.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"In the initial release of the ATB, inter-annotator agreement was inferior to other LDC treebanks (Maamouri et al., 2008).","label":0},{"source_text":"They extended previous work on discriminative weighting by using a finer granularity, focusing on the properties of instances rather than corpus components, and used simpler training procedure.","target_text":"We introduce several new ideas.","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"The second concerns the methods used (if any) to ex\u00c2\u00ad tend the lexicon beyond the static list of entries provided by the machine-readable dictionary upon which it is based.","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The precision and recall measures (described in more detail in Section 3) used in evaluating Treebank parsing treat each constituent as a separate entity, a minimal unit of correctness.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"And if one is interested in TIS, one would probably consider the single orthographic word ACL to consist of three phonological words-lei s'i d\/-corresponding to the pronunciation of each of the letters in the acronym.","label":0},{"source_text":"From a DP-based solution to the traveling salesman problem, they present a novel technique to restrict the possible word reordering between source and target language in order to achieve an e\u00c3\u0086cient search algorithm.","target_text":"Search t0 CPU time #search mWER Method [sec] error [%] QmS 0.0 0.07 108 42:6 1.0 0.13 85 37:8 2.5 0.35 44 36:6 5.0 1.92 4 34:6 10.0 10.6 0 34:5 IbmS 0.0 0.14 108 43:4 1.0 0.3 84 39:5 2.5 0.8 45 39:1 5.0 4.99 7 38:3 10.0 28.52 0 38:2 Table 6 shows example translations obtained by the three different approaches.","label":0},{"source_text":"Instance-weighting approach improved over a wide range of baselines, giving gains of over 2 BLEU points over the best non-adapted baseline.","target_text":"We used 22 features for the logistic weighting model, divided into two groups: one intended to reflect the degree to which a phrase pair belongs to general language, and one intended to capture similarity to the IN domain.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The differences in difficulty are better reflected in the BLEU scores than in the raw un-normalized manual judgements.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"36 79.","label":0},{"source_text":"The problem of coreference resolution has received considerable attention, including theoretical discourse models and supervised machine learning systems.","target_text":"However their work did not consider other types of lexical expectations (e.g., PP arguments), semantic expectations, or context comparisons like our case- frame network.(Niyu et al., 1998) used unsupervised learning to ac quire gender, number, and animacy information from resolutions produced by a statistical pronoun resolver.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"Entries for 2 < n < N store a vocabulary identifier, probability, backoff, and an index into the n + 1-gram table.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"On the MUC6 data, Bikel et al.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Moreover, if we want to test on a huge test corpus, indexing the whole corpus might prove computationally expensive.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"(Brandt 1996) extended these ideas toward a conception of kommunikative Gewichtung (\u00e2\u0080\u0098communicative-weight assignment\u00e2\u0080\u0099).","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"Two lists, Corporate-Suffix-List (for corporate suffixes) and Person-Prefix-List (for person prefixes), are collected from the training data.","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"Both parameters depend on a single hyperparameter \u00ce\u00b1.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"These knowledge sources determine whether the contexts surrounding an anaphor and antecedent are compatible.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Tables 4 and 5 also show that putting all of the contextual role KSs in play at the same time produces the greatest performance gain.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"To measure the contribution of each modification, a third, intermediate algorithm, Yarowsky-cautious was also tested.","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"On the English side, however, the vertices (denoted by Ve) correspond to word types.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"72 77.","label":0},{"source_text":"However, these approaches are ill-equipped to directly represent type-based constraints such as sparsity.","target_text":"In addition, this formulation results in a dramatic reduction in the number of model parameters thereby, enabling unusually rapid training.","label":0},{"source_text":"The PROBING data structure uses linear probing hash tables and is designed for speed.","target_text":"Timing is based on plentiful memory.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"9 65.5 46.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"The features are weighted within a logistic model to give an overall weight that is applied to the phrase pair\u2019s frequency prior to making MAP-smoothed relative-frequency estimates (different weights are learned for each conditioning direction).","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"All annotations are done with specific tools and in XML; each layer has its own DTD.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"so that 'door' would be and in this case the hanzi 7C, does not represent a syllable.","label":0},{"source_text":"Nevertheless, only a part of this corpus (10 texts), which the authors name \"core corpus\", is annotated with all this information.","target_text":"The motivation for our more informal approach was the intuition that there are so many open problems in rhetorical analysis (and more so for German than for English; see below) that the main task is qualitative investigation, whereas rigorous quantitative analyses should be performed at a later stage.","label":0},{"source_text":"This assumption, however, is not inherent to type-based tagging models.","target_text":"Specifically, the lexicon is generated as: P (T , W |\u00cf\u0088) =P (T )P (W |T ) Word Type Features (FEATS): Past unsupervised POS work have derived benefits from features on word types, such as suffix and capitalization features (Hasan and Ng, 2009; Berg-Kirkpatrick et al.,2010).","label":0},{"source_text":"The overall parsing accuracy obtained with the pseudo-projective approach is still lower than for the best projective parsers.","target_text":"The Danish Dependency Treebank (DDT) comprises about 100K words of text selected from the Danish PAROLE corpus, with annotation of primary and secondary dependencies (Kromann, 2003).","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"This alters generation of T as follows: n P (T |\u00cf\u0088) = n P (Ti|\u00cf\u0088) i=1 Note that this distribution captures the frequency of a tag across word types, as opposed to tokens.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"Automatic scores are computed on a larger tested than manual scores (3064 sentences vs. 300\u2013400 sentences). collected manual judgements, we do not necessarily have the same sentence judged for both systems (judges evaluate 5 systems out of the 8\u201310 participating systems).","label":0},{"source_text":"In this paper the author evaluates machine translation performance for six European language pairs that participated in a shared task: translating French, German, Spanish texts to English and back.","target_text":"Since the inclusion of out-ofdomain test data was a very late decision, the participants were not informed of this.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"This decreases the statistical significance of our results compared to those studies.","label":0},{"source_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"The sign test checks, how likely a sample of better and worse BLEU scores would have been generated by two systems of equal performance.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"Input: Ja , wunderbar . Konnen wir machen . MonS: Yes, wonderful.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"This logic applies recursively: if wnf+1 similarly does not extend and has zero log backoff, it too should be omitted, terminating with a possibly empty context.","label":0},{"source_text":"The contextual rules are restricted and may not be applicable to every example, but the spelling rules are generally applicable and should have good coverage.","target_text":"Many statistical or machine-learning approaches for natural language problems require a relatively large amount of supervision, in the form of labeled training examples.","label":0},{"source_text":"In this work, they take a more direct approach and treat a word type and its allowed POS tags as a primary element of the model.","target_text":"The second model (+PRIOR) utilizes the independent prior over type-level tag assignments P (T |\u00cf\u0088).","label":0},{"source_text":"One can trivially create situations in which strictly binary-branching trees are combined to create a tree with only the root node and the terminal nodes, a completely flat structure.","target_text":"The combining algorithm is presented with the candidate parses and asked to choose which one is best.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"Evaluation of Morphological Analysis.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"The maximum precision oracle is an upper bound on the possible gain we can achieve by parse hybridization.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"For our demonstration system, we typically use the pruning threshold t0 = 5:0 to speed up the search by a factor 5 while allowing for a small degradation in translation accuracy.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"These performance gains transfer to improved system runtime performance; though we focused on Moses, our code is the best lossless option with cdec and Joshua.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"92 77.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"For the error counts, a range from 0:0 to 1:0 is used.","label":0},{"source_text":"The authors show that PATB is similar to other tree-banks but that annotation consistency remains low.","target_text":"The ATB segmentation scheme is one of many alternatives.","label":0},{"source_text":"They plan on extending instance-weighting to other standard SMT components and capture the degree of generality of phrase pairs.","target_text":"The original OUT counts co(s, t) are weighted by a logistic function w\u03bb(s, t): To motivate weighting joint OUT counts as in (6), we begin with the \u201cideal\u201d objective for setting multinomial phrase probabilities 0 = {p(s|t), dst}, which is the likelihood with respect to the true IN distribution pi(s, t).","label":0},{"source_text":"they evaluate various specific aspects of the segmentation, as well as the overall segmentation performance.","target_text":"Particular instances of relations are associated with goodness scores.","label":0},{"source_text":"The authors believe automatic paraphrase discovery is an important component for building a fully automatic information extraction system.","target_text":"The data is sorted based on the frequency of the context (\u00e2\u0080\u009ca unit of\u00e2\u0080\u009d appeared 314 times in the corpus) and the NE pair instances appearing with that context are shown with their frequency (e.g. \u00e2\u0080\u009cNBC\u00e2\u0080\u009d and \u00e2\u0080\u009cGeneral Electric Co.\u00e2\u0080\u009d appeared 10 times with the context \u00e2\u0080\u009ca unit of\u00e2\u0080\u009d).","label":0},{"source_text":"There are clustering approaches that assign a single POS tag to each word type.","target_text":"This design does not guarantee \u00e2\u0080\u009cstructural zeros,\u00e2\u0080\u009d but biases towards sparsity.","label":0},{"source_text":"On the basis of that observation, they described a class of formalisms which they called Linear Context- Free Rewriting Systems (LCFRs), and showed they were recognizable in polynomial time and generated only semilinear languages.","target_text":"In order to observe the similarity between these constrained systems, it is crucial to abstract away from the details of the structures and operations used by the system.","label":0},{"source_text":"The evaluation compares the performance of the system with that of several human judges and inter-human agreement on a single correct way to segment a text.","target_text":"Table 4 Differences in performance between our system and Wang, Li, and Chang (1992).","label":0},{"source_text":"The authors cluster NE instance pairs based on the words in the context using bag-of-words methods.","target_text":"As lower frequency examples include noise, we set a threshold that an NE category pair should appear at least 5 times to be considered and an NE instance pair should appear at least twice to be considered.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"including Third Tone Sandhi (Shih 1986), which changes a 3 (low) tone into a 2 (rising) tone before another 3 tone: 'j\";gil, xiao3 [lao3 shu3] 'little rat,' becomes xiao3 { lao2shu3 ], rather than xiao2 { lao2shu3 ], because the rule first applies within the word lao3shu3 'rat,' blocking its phrasal application.","label":0},{"source_text":"In order to handle the necessary word reordering as an optimization problem within the dynamic programming approach, they describe a solution to the traveling salesman problem (TSP) which is based on dynamic programming.","target_text":"In this paper, we describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","label":0},{"source_text":"It outperformed strong unsupervised baselines as well as approaches that relied on direct projections, and bridged the gap between purely supervised and unsupervised POS tagging models.","target_text":"As one can see, only the trigrams [suo incarceramento ,], [suo iter ,] and [suo carattere ,] are connected to English words.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"Although this feature helps, we encounter one consequence of variable word order.","label":0},{"source_text":"The approach has been successfully tested on the 8 000-word Verbmobil task.","target_text":"On average, 6 reference translations per automatic translation are available.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"RandLM\u2019s stupid backoff variant stores counts instead of probabilities and backoffs.","label":0},{"source_text":"The use of global features has shown excellent result in the performance on MUC-6 and MUC-7 test data.","target_text":"Case and Zone of and : Similarly, if (or ) is initCaps, a feature (initCaps, zone) (or (initCaps, zone) ) is set to 1, etc. Token Information: This group consists of 10 features based on the string , as listed in Table 1.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"The present proposal falls into the last group.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"Our analysis identifies three key factors driving our performance gain: 1) selecting a model structure which directly encodes tag sparsity, 2) a type-level prior on tag assignments, and 3) a straightforward na\u00c2\u00a8\u00c4\u00b1veBayes approach to incorporate features.","label":0},{"source_text":"It also incorporates the Good-Turing methodin estimating the likelihoods of previously unseen constructions, including morphological derivatives and personal names.","target_text":"More complex approaches such as the relaxation technique have been applied to this problem Fan and Tsai (1988}.","label":0},{"source_text":"Here we show how non-projective dependency parsing can be achieved by combining a data driven projective parser with special graph transformation techniques.","target_text":"We also see that the increase in the size of the label sets for Head and Head+Path is far below the theoretical upper bounds given in Table 1.","label":0},{"source_text":"Here we present two algorithms.","target_text":"It is a sequence of proper nouns within an NP; its last word Cooper is the head of the NP; and the NP has an appositive modifier (a vice president at S.&P.) whose head is a singular noun (president).","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"conceptual relationship in the discourse.","label":0},{"source_text":"The TRIE model uses less memory than the smallest lossless alternative and is still faster than SRILM.","target_text":"If the key distribution\u2019s range is also known (i.e. vocabulary identifiers range from 0 to the number of words), then interpolation search can use this information instead of reading A[0] and A[|A |\u2212 1] to estimate pivots; this optimization alone led to a 24% speed improvement.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"\u2022 We evaluated translation from English, in addition to into English.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"was done by the participants.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1&#8211;5 seemed to be very hard to perform.","target_text":"The BLEU metric, as all currently proposed automatic metrics, is occasionally suspected to be biased towards statistical systems, especially the phrase-based systems currently in use.","label":0},{"source_text":"For the perplexity and translation tasks, we used SRILM to build a 5-gram English language model on 834 million tokens from Europarl v6 (Koehn, 2005) and the 2011 Workshop on Machine Translation News Crawl corpus with duplicate lines removed.","target_text":"Part of the gap between resident and virtual memory is due to the time at which data was collected.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"The test set included 2000 sentences from the Europarl corpus, but also 1064 sentences out-ofdomain test data.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"of Articles No.","label":0},{"source_text":"The bias of automatic methods in favour of statistical systems seems to be less pronounced on out-of-domain test data.","target_text":"We check, how likely only up to k = 20 better scores out of n = 100 would have been generated by two equal systems, using the binomial distribution: If p(0..k; n, p) < 0.05, or p(0..k; n, p) > 0.95 then we have a statistically significant difference between the systems.","label":0},{"source_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold.","target_text":"The most frequent NE category pairs are \u00e2\u0080\u009cPerson - Person (209,236), followed by \u00e2\u0080\u009cCountry - Coun- try\u00e2\u0080\u009d (95,123) and \u00e2\u0080\u009cPerson - Country\u00e2\u0080\u009d (75,509).","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"In order to create good-sized vectors for similarity calculation, they had to set a high frequency threshold, 30.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","label":0},{"source_text":"They focused on phrases which two Named Entities, and proceed in two stages.","target_text":"If there is a frequent multi-word sequence in a domain, we could use it as a keyword candidate.","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"shows some keywords with their scores.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"to represent the ith word type emitted by the HMM: P (t(i)|Ti, t(\u00e2\u0088\u0092i), w, \u00ce\u00b1) \u00e2\u0088\u009d n P (w|Ti, t(\u00e2\u0088\u0092i), w(\u00e2\u0088\u0092i), \u00ce\u00b1) (tb ,ta ) P (Ti, t(i)|T , W , t(\u00e2\u0088\u0092i), w, \u00ce\u00b1, \u00ce\u00b2) = P (T |tb, t(\u00e2\u0088\u0092i), \u00ce\u00b1)P (ta|T , t(\u00e2\u0088\u0092i), \u00ce\u00b1) \u00e2\u0088\u0092i (i) i i (\u00e2\u0088\u0092i) P (Ti|W , T \u00e2\u0088\u0092i, \u00ce\u00b2)P (t |Ti, t , w, \u00ce\u00b1) All terms are Dirichlet distributions and parameters can be analytically computed from counts in t(\u00e2\u0088\u0092i)where T \u00e2\u0088\u0092i denotes all type-level tag assignment ex cept Ti and t(\u00e2\u0088\u0092i) denotes all token-level tags except and w (\u00e2\u0088\u0092i) (Johnson, 2007).","label":0},{"source_text":"Most IE researchers have been creating paraphrase knowledge by hand and specific tasks.","target_text":"As you can see in the figure, the accuracy for the domain is quite high except for the \u00e2\u0080\u009cagree\u00e2\u0080\u009d set, which contains various expressions representing different relationships for an IE application.","label":0},{"source_text":"All the texts were annotated by two people.","target_text":"This was also inspired by the work on the Penn Discourse Tree Bank7 , which follows similar goals for English.","label":0},{"source_text":"Other kinds of productive word classes, such as company names, abbreviations,and place names can easily be handled given appropriate models.","target_text":"Making the reasonable assumption that similar information is relevant for solving these problems in Chinese, it follows that a prerequisite for intonation-boundary assignment and prominence assignment is word segmentation.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"It may be more realistic to replace the second criteria with a softer one, for example (Blum and Mitchell 98) suggest the alternative Alternatively, if Ii and 12 are probabilistic learners, it might make sense to encode the second constraint as one of minimizing some measure of the distance between the distributions given by the two learners.","label":0},{"source_text":"The resulting model is compact, efficiently learnable and linguistically expressive.","target_text":"5 67.3 55.","label":0},{"source_text":"It is probably the first analysis of Arabic parsing of this kind.","target_text":"(2006) developed a technique for splitting and chunking long sentences.","label":0},{"source_text":"This paper offers a broad insight into of Arabic constituency parsing by analyzing the interplay of linguistic phenomena, annotation choices, and model design.","target_text":"Evalb, the standard parsing metric, is biased toward such corpora (Sampson and Babarczy, 2003).","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"We can better predict the probability of an unseen hanzi occurring in a name by computing a within-class Good-Turing estimate for each radical class.","label":0},{"source_text":"The overall performance of their joint framework demonstrated that a probability distribution obtained over mere syntactic contexts using a Treebank grammar and a data-driven lexicon outperformed upper bounds proposed by previous joint disambiguation systems and achieved segmentation and parsing results on a par with state-of-the-art standalone applications results.","target_text":"Lexical rules are estimated in a similar manner.","label":0},{"source_text":"The approach assumes that the word reordering is restricted to a few positions in the source sentence.","target_text":"However, dynamic programming can be used to find the shortest tour in exponential time, namely in O(n22n), using the algorithm by Held and Karp.","label":0},{"source_text":"The manual evaluation of scoring translation on a graded scale from 1\u00e2\u0080\u00935 seems to be very hard to perform.","target_text":"Given a set of n sentences, we can compute the sample mean x\ufffd and sample variance s2 of the individual sentence judgements xi: The extend of the confidence interval [x\u2212d, x+df can be computed by d = 1.96 \u00b7\ufffdn (6) Pairwise Comparison: As for the automatic evaluation metric, we want to be able to rank different systems against each other, for which we need assessments of statistical significance on the differences between a pair of systems.","label":0},{"source_text":"This paper presents a maximum entropy-based named entity recognizer (NER).","target_text":"Global features are extracted from other occurrences of the same token in the whole document.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"(Brandt 1996) extended these ideas toward a conception of kommunikative Gewichtung (\u00e2\u0080\u0098communicative-weight assignment\u00e2\u0080\u0099).","label":0},{"source_text":"Their results show that their high performance NER use less training data than other systems.","target_text":"Ltd., then organization will be more probable.","label":0},{"source_text":"They showed the efficacy of graph-based label propagation for projecting part-of-speech information across languages.","target_text":"We use N(u) to denote the neighborhood of vertex u, and fixed n = 5 in our experiments.","label":0},{"source_text":"The authors in this paper describe a search procedure for statistical machine translation (MT) based on dynamic programming (DP).","target_text":"In the second and third translation examples, the IbmS word reordering performs worse than the QmS word reordering, since it can not take properly into account the word reordering due to the German verbgroup.","label":0},{"source_text":"The code is open source, has minimal dependencies, and offers both C++ and Java interfaces for integration.","target_text":"The PROBING model was designed to improve upon SRILM by using linear probing hash tables (though not arranged in a trie), allocating memory all at once (eliminating the need for full pointers), and being easy to compile.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"Therefore, we want state to encode the minimum amount of information necessary to properly compute language model scores, so that the decoder will be faster and make fewer search errors.","label":0},{"source_text":"They have made use of local and global features to deal with the instances of same token in a document.","target_text":"This group contains a large number of features (one for each token string present in the training data).","label":0},{"source_text":"Das and Petrov, in this paper, approached inducing unsupervised part-of-speech taggers for languages that had no labeled training data, but had translated text in a resource-rich language.","target_text":"The edge weights between the foreign language trigrams are computed using a co-occurence based similarity function, designed to indicate how syntactically similar the middle words of the connected trigrams are (\u00a73.2).","label":0},{"source_text":"Replacing this with a ranked evaluation seems to be more suitable.","target_text":"However, a recent study (Callison-Burch et al., 2006), pointed out that this correlation may not always be strong.","label":0},{"source_text":"Across eight European languages, their approach results in an average absolute improvement of 10.4% over a state-of-the-art baseline, and 16.7% over vanilla hidden Markov models induced with the Expectation Maximization algorithm.","target_text":"Because all English vertices are going to be labeled, we do not need to disambiguate them by embedding them in trigrams.","label":0},{"source_text":"Throughout this paper we compare with several packages: SRILM 1.5.12 (Stolcke, 2002) is a popular toolkit based on tries used in several decoders.","target_text":"The TRIE model continues to use the least memory of ing (-P) with MAP POPULATE, the default.","label":0},{"source_text":"The departure from the traditional token-based tagging approach allow them to explicitly capture type-level distributional properties of valid POS tag assignments as part of the model.","target_text":"We observe similar trends when using another measure \u00e2\u0080\u0093 type-level accuracy (defined as the fraction of words correctly assigned their majority tag), according to which La ng ua ge M etr ic B K 10 E M B K 10 L B F G S G 10 F EA T S B es t F EA T S M ed ia n E ng lis h 1 1 m 1 4 8 . 3 6 8 . 1 5 6 . 0 7 5 . 5 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 0 . 9 6 6 . 4 4 7 . 8 6 6 . 4 D an is h 1 1 m 1 4 2 . 3 6 6 . 7 4 2 . 6 5 8 . 0 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 2 . 1 6 1 . 2 4 3 . 2 6 0 . 7 D ut ch 1 1 m 1 5 3 . 7 6 7 . 0 5 5 . 1 6 4 . 7 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 6 . 4 6 9 . 0 5 1 . 5 6 7 . 3 Po rtu gu es e 1 1 m 1 5 0 . 8 7 5 . 3 4 3 . 2 7 4 . 8 44 .5 69 .2 6 4 . 1 7 4 . 5 5 6 . 5 7 0 . 1 S pa ni sh 1 1 m 1 \u00e2\u0080\u0093 \u00e2\u0080\u0093 4 0 . 6 7 3 . 2 \u00e2\u0080\u0093 \u00e2\u0080\u0093 5 8 . 3 6 8 . 9 5 0 . 0 5 7 . 2 Table 4: Comparison of our method (FEATS) to state-of-the-art methods.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"For a description of the application of AdaBoost to various NLP problems see the paper by Abney, Schapire, and Singer in this volume.","label":0},{"source_text":"Through parser combination we have reduced the precision error rate by 30% and the recall error rate by 6% compared to the best previously published result.","target_text":"There is a guarantee of no crossing brackets but there is no guarantee that a constituent in the tree has the same children as it had in any of the three original parses.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"This is motivated by taking \u03b2 po(s|t) to be the parameters of a Dirichlet prior on phrase probabilities, then maximizing posterior estimates p(s|t) given the IN corpus.","label":0},{"source_text":"This paper talks about Exploiting Diversity in Natural Language Processing: Combining Parsers.","target_text":"The computation of Pfr1(c)1Mi M k (C)) has been sketched before in Equations 1 through 4.","label":0},{"source_text":"Vijay-Shankar et all considered the structural descriptions produced by various grammatical formalisms in terms of the complexity of the paths and the relationship between paths in the sets of structural descriptions that each system can generate.","target_text":"The operations must be linear and nonerasing, i.e., they can not duplicate or erase structure from their arguments.","label":0},{"source_text":"It is annotated with several data: morphology, syntax, rhetorical structure, connectors, correference and informative structure.","target_text":"Either save money at any cost - or give priority to education.","label":0},{"source_text":"They proposed a single joint model for performing both morphological segmentation and syntactic disambiguation which bypasses the associated circularity.","target_text":"Evaluation We use 8 different measures to evaluate the performance of our system on the joint disambiguation task.","label":0},{"source_text":"Each out-of-domain phrase pair was characterized by a set of simple features intended to reflect how useful it would be.","target_text":"The logistic function, whose outputs are in [0, 1], forces pp(s, t) <_ po(s, t).","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"We showed in \u00c2\u00a72 that lexical ambiguity explains the underperformance of these categories.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"All three curves remain steep at the maximum training set size of 18818 trees.","label":0},{"source_text":"They believe that global context is useful in most languages, as it is a natural tendency for authors to use abbreviations on entities already mentioned previously.","target_text":"A secondary reference resolution classifier has information on the class assigned by the primary classifier.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"For more on the participating systems, please refer to the respective system description in the proceedings of the workshop.","label":0},{"source_text":"A large number of rules are needed for coverage of the domain, suggesting that a fairly large number of labeled examples should be required to train a classifier.","target_text":"The input to AdaBoost is a set of training examples ((xi , yi), , (x\u201e.\u201e yrn)).","label":0},{"source_text":"There is no global pruning.","target_text":"On average, 6 reference translations per automatic translation are available.","label":0},{"source_text":"The experimental tests are carried out on the Verbmobil task which is a limited-domain spoken-language task.","target_text":"Subsets of partial hypotheses with coverage sets C of increasing cardinality c are processed.","label":0},{"source_text":"These clusters are computed using an SVD variant without relying on transitional structure.","target_text":"9 50.2 +P RI OR be st me dia n 47.","label":0},{"source_text":"Bean and Riloff also used bootstrapping to extend their semantic compatibility model, proposed using caseframe network for anaphora resolution, information extraction patterns to identify contextual clues for determining compatibility between NPs.","target_text":"Our representation of contextual roles is based on information extraction patterns that are converted into simple caseframes.","label":0},{"source_text":"The first method builds on results from (Yarowsky 95) and (Blum and Mitchell 98).","target_text":"Our derivation is slightly different from the one presented in (Schapire and Singer 98) as we restrict at to be positive.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"However, it is possible to personify any noun, so in children's stories or fables, i\u00c2\u00a5JJ1l.","label":0},{"source_text":"The AdaBoost algorithm was developed for supervised learning.","target_text":"We make the assumption that for each example, both xi,. and x2,2 alone are sufficient to determine the label yi.","label":0},{"source_text":"This architecture provides a uniform framework in which it is easy to incorporate not only listed dictionary entries but also morphological derivatives, and models for personal names and foreign names in transliteration.","target_text":"Mandarin exhibits several such processes, including A-not-A question formation, il\u00c2\u00ad lustrated in (3a), and adverbial reduplication, illustrated in (3b): 3.","label":0},{"source_text":"Koehn and Monz carried out an extensive manual and automatic evaluation of machine translation performance on European language pairs.","target_text":"The judgement of 4 in the first case will go to a vastly better system output than in the second case.","label":0},{"source_text":"It is well-known that English constituency parsing models do not generalize to other languages and treebanks.","target_text":"Table 9: Dev set results for sentences of length \u00e2\u0089\u00a4 70.","label":0},{"source_text":"In order to realize the full potential of dependency-based syntactic parsing, it is desirable to allow non-projective dependency structures.","target_text":"From the point of view of computational implementation this can be problematic, since the inclusion of non-projective structures makes the parsing problem more complex and therefore compromises efficiency and in practice also accuracy and robustness.","label":0},{"source_text":"In this paper, the authors proposed an approach for instance-weighting phrase pairs in an out-of-domain corpus in order to improve in-domain performance.","target_text":"We model po(s|t) using a MAP criterion over weighted phrase-pair counts: and from the similarity to (5), assuming y = 0, we see that w\u03bb(s, t) can be interpreted as approximating pf(s, t)\/po(s, t).","label":0},{"source_text":"While many systems had similar performance, the results offered interesting insights, especially, about the relative performance of statistical and rule-based systems.","target_text":"The judgements tend to be done more in form of a ranking of the different systems.","label":0},{"source_text":"In this paper the authors present a stochastic finite-state model for segmenting Chinese text into words.","target_text":"then define the best segmentation to be the cheapest or best path in Id(I) o D* (i.e., Id(I) composed with the transitive closure of 0).6 Consider the abstract example illustrated in Figure 2.","label":0},{"source_text":"The model incorporates various recent techniques for incorporating and manipulating linguistic knowledge using finite-state transducers.","target_text":"Their results are then compared with the results of an automatic segmenter.","label":0},{"source_text":"This corpus has several advantages: it is annotated at different levels.","target_text":"Reiche\u00e2\u0080\u0099s colleagues will make sure that the concept is waterproof.","label":0},{"source_text":"The authors use linguistic and annotation insights to develop a manually annotated grammar and evaluate it and finally provide a realistic evaluation in which segmentation is performed in a pipeline jointly with parsing.","target_text":"Coverage indicates the fraction of hypotheses in which the character yield exactly matched the reference.","label":0},{"source_text":"An extended lexicon model is defined, and its likelihood is compared to a baseline lexicon model, which takes only single-word dependencies into account.","target_text":"The approach recursively evaluates a quantity Q(C; j), where C is the set of already visited cities and sj is the last visited city.","label":0},{"source_text":"However, using the top-level semantic classes of WordNet proved to be problematic as the class distinctions are too coarse.","target_text":"During coreference resolution, the caseframe network provides evidence that an anaphor and prior noun phrase might be coreferent.","label":0},{"source_text":"Evaluation was done automatically using the BLEU score and manually on fluency and adequacy.","target_text":"Human judges also pointed out difficulties with the evaluation of long sentences.","label":0}]